Title: Day-0 Pro activation writes no server-side outcome row, so the failing cohort is invisible in Convex

Body:

Split out of #5600 (blind spot 3 of 4). The other three closing artifacts shipped in the #5600 PR:
failure-vs-skip telemetry, browser-error Sentry capture, and the bounded entitlement marker.

The gap

persistActivationOutcomeWithRetry returns early on the day-0 path:

// src/components/ProActivationInterstitial.ts
if (!options.onlyIfUnactivated || !options.expectedActivationKey || !options.activationClaimNonce) {
  return;
}

onlyIfUnactivated: false (the day-0, post-checkout path) leaves expectedActivationKey and
activationClaimNonce undefined, so claimProActivationPresentation,
confirmProActivationPresentation, and recordProActivationOutcome are all skipped. The
proActivationPresentations table and report-activation-lift.mjs therefore cover only the
retro/backfill cohort — the one that was working.

Why it mattered

During the #5600 investigation, the day-0 cohort could only be reconstructed by joining Umami
sessions against Convex notificationChannels/alertRules rows and inferring absence. A
server-side outcome row would have made "5/5 day-0 sessions failed every write attempt" a single
query instead of a multi-source forensic exercise.

Why it was deferred

The retro path's ledger is guarded by a claim/nonce lease (claimProActivationPresentation →
confirmProActivationPresentation → recordProActivationOutcome) that exists to make the
markerless backfill fire exactly once per subscription. The day-0 path has no equivalent identity
to key on, so this is a design task, not a wiring task — it needs a decision about what the day-0
presentation key is and how it interacts with the existing already_presented / already_claimed
re-claim guards (see #5584/#5590 for the failure mode when that ordering is wrong).

Acceptance

• A day-0 activation session produces a proActivationPresentations row with its step outcomes.
• The row cannot block a legitimate later retro re-claim for the same subscription.
• report-activation-lift.mjs can separate the day-0 and retro cohorts.
• Regression coverage in e2e/pro-activation.spec.ts (the markerless flow harness already exists).