v0.3.0 - Queue status, drain mode, stack traces and variant replay

Brings the fixes from the bugfix-combined-fixes branch to main.

Changes

• Queue status names its source (#4)
  • nodeindexqueue:status labels the counts with the backend's own state names and shows the backend class and table it counted, plus a total.
  • It says that "ready" includes jobs released for a retry.
  • An unreachable queue now prints a readable error instead of a stack trace.
• Drain mode for the worker (#4)
  • nodeindexqueue:work --exit-when-empty stops about a second after nothing more can be reserved.
  • --limit keeps its meaning, so a supervised daemon still waits for new work.
  • Every exit prints the remaining queue depth.
• Stack traces for failed jobs (#3)
  • The first failure of a streak is stored through Flow's ThrowableStorage: an exception dump, or a Sentry event when Sentry is installed.
  • Repeated failures of the same streak stay one line each.
• Variant repairs survive moves and deletions (#2)
  • Runtime indexing and repair jobs capture the fulltext root and the target dimension combinations when they are queued, and replay replaceVariants() against live.
  • The worker clears content repository and Doctrine read caches before each job.
  • The Meilisearch write buffer is flushed after every indexing or removal job.

Upgrading

• Still requires medienreaktor/meilisearch dev-bugfix/combined-fixes.
• From v0.2.x: drain the live queue before deploying and restart the workers afterwards. If the queue cannot be drained, flush the index and rebuild it once after the deploy (see README).
• From dev-bugfix-combined-fixes at 6bb76c0: only #4 is new. Queued jobs are unchanged, so no drain or rebuild is needed.