…to a falsy value

EENGINE_FEATURE_* variables with values like "false" or "0" called the
nonexistent Set.remove() and threw a TypeError while loading the module.
Use Set.delete() and anchor the truthy value pattern so that values like
"latest" no longer enable a flag by accident.

The outbox API doubled the actual BullMQ retry delay by using
2^attemptsMade instead of 2^(attemptsMade-1) for jobs whose attempt
counter already includes the failed attempt. The submit worker had the
correct multiplier but advertised a retry time on the final attempt and
serialized exhausted retries as the 1970 epoch. Also removed a dead
exhaustion re-check in handleSubmitError that referenced a nonexistent
jobData.opts field and prevented the feedbackKey failure update.

Accounts added while no license was active stayed in the initializing
state forever because account assignment was only triggered by the
worker crash recovery path, not when suspended IMAP workers were
respawned after a license became active.

… ci]

- release.yaml: also tag releases as :latest on GHCR, matching Docker Hub
- deploy.yml: publish per-push dev images as :edge instead of :latest, so
  :latest is owned solely by the release workflow
- deploy.yml: pass EE_COMMIT_HASH=$GITHUB_SHA to update-info.sh
- getswagger.sh: refuse to start if the port is already in use, bound curl
  requests with --max-time, wait for the server to exit during cleanup
- update-info.sh: only trust git rev-parse when the repository toplevel is
  the working directory, so a source export nested inside an unrelated git
  repository does not record the parent repository's commit hash

…ed API bugs

Audit of the Joi response schemas across all API route modules so the
generated OpenAPI spec matches what the handlers actually return, plus
fixes for bugs the audit uncovered:

- return 404 (MessageNotFound/FolderNotFound) instead of a false response
  body when a message or folder is not found (Account#assertFound)
- resolve special-use folder aliases (\Sent, \Trash, ...) for IMAP bulk
  update, move, and delete operations
- Gmail API: support flags.set, reject unsupported labels.set with 400
  before any API calls, return 404 for unknown folder paths in bulk updates
- flags.set now takes precedence over flags.add/delete on Gmail and
  Outlook accounts, matching the IMAP backend behavior
- Outlook: only force-delete messages in bulk delete when the source
  folder is the Trash folder; report a missing Trash folder as 404
- oauth-token endpoint returns the provider name and the OAuth2 app ID as
  separate fields instead of the app ID in the provider field
- encrypt mailRuClientSecret, imapProxyServerPassword, and serviceKey
  settings at rest and mask them in GET /v1/settings responses
- OAuth2 app updates only map the deprecated tenant field into authority
  when explicitly selected via authority='tenant'
- accounts can be filtered by state 'paused' (API and admin UI)
- enforce the documented force=true query parameter on the template flush
  endpoint (DELETE /v1/templates/account/{account} without it now fails
  with 400 ForceRequired)
- assorted fixes: trackOpens flag, custom submit headers, outbox created
  timestamps, token listing IDs, delivery-test SPF comment cleanup, search
  size filters, OAuth2 app listing pagination totals, gateway delete
  response, queue stats resilience, chat success flag semantics

Known gaps left for follow-up: IMAP move target paths do not resolve
special-use aliases, Gmail bulk label updates do not map special-use
names, and the OAuth2 create route still accepts the inert tenant field.

…p ci]

jsxgettext bundled acorn 5 and crashed with a SyntaxError on any file using
post-ES2018 syntax (optional chaining etc.), which forced a hardcoded list of
scanned files and a ban on modern syntax in them.

The new gettext-extract.js parses with acorn 8 (ecmaVersion latest), walks the
AST for gettext()/ngettext() calls, and merges the strings into
translations/messages.pot via gettext-parser. Files are discovered dynamically
under bin/, lib/, and workers/, so new modules with translatable strings are
picked up automatically - no list to maintain when routes are refactored.

The extracted msgid set is identical to the previous output, the views half of
the pipeline (xgettext-template) is unchanged, and the "no newer syntax"
restriction comments were removed from the previously scanned files since the
constraint no longer applies.

Continues the response schema audit: adds response schemas with
failAction 'log' to the remaining API routes (blocklist, chat,
delivery-test, export, gateway, license, mailbox, message, oauth2-app,
outbox, pubsub, settings, stats, submit, template, token, webhook-route),
extends lib/schemas.js with the shared error response helper usage, and
simplifies the gettext coverage test.

The unquoted lib/**/*.js globs in the lint script and Gruntfile were
expanded by sh (no globstar) to depth-2 files only, so the 48 root-level
lib files and everything three or more levels deep were never linted by
npm test or CI. Quoting the globs lets eslint expand them itself.

Fixes the issues this surfaced (all behavior-neutral):

- lib/account.js: drop a dead initializer (no-useless-assignment)
- lib/autodetect-imap-settings.js: remove a leftover dead filter statement
- lib/email-client/imap/subconnection.js: remove a no-op null assignment
- lib/email-client/imap/sync-operations.js, lib/stream-encrypt.js:
  attach the caught error as cause when rethrowing (preserve-caught-error)
- lib/imapproxy/imap-core/lib/imap-connection.js: remove a dead assignment
- lib/imapproxy/imap-core/test/imap-indexer-test.js: keep the smoke parse
  without the unused assignment
- lib/metrics-collector.js, lib/redis-operations.js: drop unused variables

The tenth surfaced error (lib/webhooks.js:401) was a real bug and is
fixed separately.

…s, and schemas

Bugs:

- Gmail API: resolve add/remove label conflicts in favor of the add in
  updateMessage/updateMessages, so bulk deletes from the Trash folder
  (add TRASH + remove TRASH) and moves within the current folder no
  longer fail with 400, and map IMAP special-use labels to Gmail label
  IDs in bulk updates via a shared toGmailLabelId helper (single-message
  updates already mapped them)
- export worker: isSkippableError now reads the machine-readable code
  from Boom output payloads (like its sibling isFolderMissingError), so
  a message expunged mid-export is skipped instead of failing the whole
  export job; the error classifiers moved to lib/export.js for reuse
  and test coverage
- webhooks: the handler cache version marker was assigned backwards and
  the per-route version compared a number to the string from Redis, so
  the cache was rebuilt from Redis on every webhook event once a custom
  route existed; the cache now refreshes only when routes change
- settings: enabling generateEmailSummary or openAiGenerateEmbeddings
  now really enables notifyText - the old 'key in [array]' condition
  never matched - while still storing the key itself and bumping the
  OpenAI settings version
- schemas: token access.ip and the short mailbox listing delimiter
  allow null, remoteAddress is documented as the IP of the client that
  created the token (not a restriction), and search responses document
  the documentStoreQuery debug field
- gettext-extract: skip empty msgids that would corrupt the POT header
  entry

Cleanup: shared mailbox base fields and a verify-result factory in the
response schemas, shared OAuth2 scope-array schema, hoisted template
account field, single schemas require in oauth2-app-routes, dropped the
serviceUrl normalization from the settings route (settings.set already
normalizes via formatSettingValue), removed unused schema exports and
the dead module.exports block in workers/export.js.

Tests: new regression coverage for the export error classifiers, the
webhook handler cache, the settings AI/notifyText coupling, and gettext
extraction (interpolated template literal tripwire, ngettext-only
module coverage, empty msgid guard). translations/messages.pot
regenerated for the final line references.

Error tracking now uses @sentry/node gated on the SENTRY_DSN env var,
replacing @bugsnag/js and BUGSNAG_API_KEY. The nine duplicated init
blocks in server.js and the worker entry points are consolidated into
lib/sentry.js. Sentry is configured for error capture only: the SDK is
required lazily so disabled installs pay no startup cost, and default
integrations plus OpenTelemetry setup are skipped to keep http/fetch
hot paths unpatched.

The global uncaughtException/unhandledRejection handlers in lib/logger.js
now always terminate the process, reporting the error and waiting for a
short delivery flush first when tracking is enabled. Previously exit was
delegated to the tracking SDK, which Sentry does not guarantee in worker
threads.

Adds sentryEnabled and sentryDsn settings, exposed on the Config ->
Logging page and through /v1/settings. When enabled without a custom
DSN, reports go to the shared community Sentry instance run by the
EmailEngine developers, so error reporting can be switched on
temporarily during a support case without setting up a Sentry server.

The settings are applied at runtime: lib/sentry.js polls them once a
minute in every thread and starts or stops the SDK on change, so no
restart is needed. The SENTRY_DSN environment variable still pins the
configuration and disables the runtime settings.

Events from different EmailEngine installations sharing the community
Sentry DSN are now attributable: every report carries an 'instance' tag
(the unique serviceId of the installation) and a 'license' tag (the
license key id, read from the cached license metadata). Identity values
are read once per SDK start, so a license swapped at runtime shows up
after the next restart. A failed identity lookup only means untagged
events, it never breaks error reporting.

…er--components--emailengine-app

chore(master): release 2.70.0