fix(sync): drop durable calendar-list discovery failures

[cursor]

Summary

Stops Google calendar-list durable 4xx refusals (e.g. "The user must be signed up for Google Calendar.") from burning the sync retry ladder and tripping self-heal alarms.

PostHog cluster (7d):

• Sync job engine failed… — 159 occ; 79 were calendarListSync on connection 6a653974… with root cause ProviderCalendarError(discoveryFailed): … The user must be signed up for Google Calendar.
• Sync self-heal sweep… — 39 occ; exhausted requeue budget for the same job 6a76454e…
• Prior subscriptionMaintain / watchFailed half of the same fingerprint already fixed by fix(sync): settle durable google watch failures instead of retrying #2635

Root cause: google-calendar.adapter mapped every non-410 list failure to discoveryFailed, and syncCalendarList / dispatch treated that as transient → retryableTransient × 20 → failed → self-heal requeue loop.

Fix (mirrors event readFailed / watch watchFailed):

• Classify Google calendarList errors: cursorExpired (410), transient (429/5xx/network + rate-limit 403 reasons), discoveryFailed (durable 4xx)
• Enrich cause with HTTP status + Google reason for triage
• dispatchSyncJob drops durable discoveryFailed, stamps lastReadFailure* on the calendarList resource, frees the coalescing key for later rediscovery/reconnect

Simplicity

Minimal path parallel to existing readFailed settlement — no new job failure class, no health-model redesign. Classification helpers stay local to the Google calendar adapter (same shape as event-reader / notifications).

Automated validation

• bun test:sync -- packages/sync/src/providers/google/google-calendar.adapter.test.ts packages/sync/src/domain/sync-job-dispatch.service.db.test.ts packages/sync/src/domain/calendar-list-sync.service.db.test.ts → 48 pass
• bun run lint → no new issues (existing repo warnings only)

Independent review

Self-reviewed against the prod log cause chain and the settled readFailed / watchFailed patterns. Confirmed rate-limit 403s remain transient.

Test plan

• bun test:sync -- packages/sync/src/providers/google/google-calendar.adapter.test.ts packages/sync/src/domain/sync-job-dispatch.service.db.test.ts packages/sync/src/domain/calendar-list-sync.service.db.test.ts
• bun run lint

Remaining risk / follow-up

• After deploy, clear or revive exhausted failed job 6a76454e6a9f4b8649b55e41 (calendarListSync:6a6539741eaf145653f64365) so rediscovery can re-enqueue; self-heal budget is already exhausted for that row.
• Backend GenericError.NotSure (“Not sure why error occurred. See logs”, 60 occ) from calendar list proxy remains opaque in PostHog because error() puts the useful kind on result, not Error.message — separate triage improvement, not required for this storm.
• Resolve PostHog issues 019fd9ff-… / 019fdc65-… once volume stops post-deploy.