Adds support for FlexLib v4.2.18's NAVTEX maritime broadcast waveform.
Scope is data layer + protocol plumbing only, per #2132's explicit
"visual design + UX is maintainer-only" instruction.

Carved out of bot PR #2137; the standalone NavtexApplet UI class it
included was unwired (no AppletPanel registration, no construction
site in MainWindow, no activate/deactivate command path), so it would
have been dead code in production.  Leaves the UI work for a
maintainer-driven follow-up when the activate/deactivate UX is
decided.  See PR #2137 review comment for the full rationale.

What's included:

- NavtexModel: parses `navtex` and `navtex sent` status messages,
  formats `navtex send tx_ident=... subject_indicator=... msg_text=...`,
  tracks per-message state (Pending → Queued → Sent / Error).
  Pending messages keyed by command sequence number; promoted to
  Queued via the reply handler then to Sent via the broadcast.

- RadioModel wiring: subscribes to `sub navtex all` and routes status
  to NavtexModel.  Forwards commandReady and replyCommandReady, with
  a reply handler that drives handleSendResponse.

- SliceModel: NT mode classified as USB-family for filter/AGC purposes
  (matches FlexLib's slice-mode handling for digital data modes).

- MainWindow / RxApplet / VfoWidget: NT mode threaded through the
  mode-string switch so the GUI doesn't crash or fall through to
  "unknown" when a slice enters mode NT.

Improvements over the original PR:

1. Quote-escape `msg_text="..."` in NavtexModel::sendMessage.  The
   original `arg(msgText)` substitution would break the radio's command
   parser if the user's message contained a `"` or `\`.  Backslash is
   escaped first to avoid double-escaping existing backslashes.

2. Failure-path bug fix in handleSendResponse: failure responses come
   as `R<seq>|<errcode>|` with empty body, so the original "if
   indexStr empty bail" check fired BEFORE the "if respVal != 0 mark
   Error" branch — making error reporting unreachable.  Fixed by
   reordering: check pending entry first, then respVal != 0, then
   index parsing.

3. New unit test `tests/navtex_model_test.cpp` (21 assertions)
   covering: Pending → Queued promotion, Queued → Sent promotion,
   send-failure path, orphan navtex-sent (Multi-Flex case), case-
   insensitive status string parsing, msg_text quote/backslash
   escaping, idempotent navtex-sent handling.

Fixes #2132.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

The DxClusterDialog::freedvReportingToggled connect lambda references
m_radeEngine, m_radeSliceId, startFreeDvReporting(), and stopFreeDvReporting()
— all of which are only declared under #ifdef HAVE_RADE. The connect
itself was wrapped in #ifdef HAVE_WEBSOCKETS only, so on Windows
(WebSockets present, RADE absent) the build broke with C2065/C3861.

Introduced in #2173 (FreeDV Reporter), but check-windows skipped that
PR because it didn't touch CMakeLists.txt / third_party / workflows.
This NAVTEX branch modifies CMakeLists.txt for the new test executable,
so check-windows fired and surfaced the regression.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>