Add retry with backoff for failed publish and update cron workers

[pfefferle]

Fixes #179
Fixes #181

Proposed changes:

• Retry with backoff for failed publish/update cron workers (Retry/backoff for failed post-publish cron events #179). Publishing is deferred to one-shot wp_schedule_single_event cron events; a single transient PDS failure (429/5xx/timeout) or missed cron tick previously dropped the post silently — the post showed "Published" in WordPress but never reached Bluesky (reported on wp.org as "inconsistent" publishing). Failed atmosphere_publish_post / atmosphere_update_post workers now re-queue themselves with an escalating 60s/5m/15m ladder, tracked in _atmosphere_publish_retries post meta and cleared on success, permanent rejection, exhaustion, or a fresh save. Mirrors the existing comment parent-defer pattern and the ActivityPub dispatcher's filterable-retry convention.
• Transient-vs-permanent classification. Retry-by-default with a permanent blocklist: local preconditions (not_connected, needs_reauth, decrypt, did_mismatch, …) and deterministic PDS 4xx are never retried; no-status network failures, 408, 429, and 5xx are. atmosphere_thread_rollback_failed is explicitly permanent — retrying after a failed rollback would mint fresh TIDs next to live orphan records and publish a duplicate copy.
• New atmosphere_publish_retry_delays filter. One entry per retry, so the ladder's length is also the attempt budget — an empty array disables retries, a longer array raises the cap. Documented in the hook reference.
• Fix: disconnect/uninstall cron cleanup missed arg-carrying events (Disconnect/uninstall cron cleanup misses events scheduled with args #181). wp_clear_scheduled_hook( $hook ) with no args only removes events scheduled with an empty args array, so every queued per-post event and the revoke event (encrypted token payload in its args) survived clear_scheduled_hooks() — the exact events the cleanup exists to stop from firing against a different connection. Both helpers now use wp_unschedule_hook().

Other information:

• Have you written new tests for your changes, if applicable?

Reviewed pre-PR by three parallel review passes (correctness/WP-Cron races, conventions/test hygiene, edge cases/abuse); all blocking findings addressed in the last commit (duplicate-post guard, stranded-counter reset, uninstall sweep).

Testing instructions:

• Start the environment: npm run env-start, connect a Bluesky account, and enable auto-publish.
• Simulate a transient PDS failure: add_filter( 'atmosphere_pre_apply_writes', fn() => new \WP_Error( 'atmosphere_pds', 'boom', array( 'status' => 500 ) ) ); (e.g. via a mu-plugin), then publish a post.
• Check wp cron event list (or WP Crontrol): an atmosphere_publish_post event for the post is queued ~60s out; after it fails again, ~5m; then ~15m; then it stops. The post's _atmosphere_publish_retries meta tracks the attempt count and is removed at the end.
• Remove the failure filter before the next retry fires: the post publishes and the counter is cleared.
• Repeat with array( 'status' => 400 ): no retry is scheduled.
• For Disconnect/uninstall cron cleanup misses events scheduled with args #181: publish a post so a per-post event is queued (or schedule one manually), then disconnect the account on the settings page and confirm wp cron event list shows no remaining atmosphere_* events.
• Run npm run env-test — 848 tests pass, including 10 new ones covering the ladder, the filter, the rollback guard, the counter reset, and the arg-carrying cleanup.

[Copilot]

Pull request overview

Adds resilience to Atmosphere’s WP-Cron publish/update workers by introducing bounded retry-with-backoff for transient failures, and fixes disconnect/deactivation/uninstall cleanup so arg-carrying cron events are reliably removed (preventing queued tasks from firing under a new connection).

Changes:

• Add transient-vs-permanent failure classification for publish/update worker errors, and reschedule transient failures with a filterable backoff ladder tracked in _atmosphere_publish_retries.
• Fix scheduled-hook cleanup by switching from wp_clear_scheduled_hook() (arg-sensitive) to wp_unschedule_hook() (args-agnostic), including coverage for revoke events with args.
• Add PHPUnit coverage for retry scheduling/escalation/exhaustion, rollback non-retry, retry-budget reset, and cron cleanup behavior; update docs and changelog entries.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
includes/class-atmosphere.php	Implements publish/update retry-with-backoff logic, retry classification, and retry-budget reset on status transition.
includes/functions.php	Fixes cron cleanup to unschedule hooks regardless of args; ensures revoke event cleanup is also args-agnostic.
tests/phpunit/tests/class-test-atmosphere.php	Adds tests validating retry ladder behavior, filter behavior, non-retryable rollback failure, and counter reset.
tests/phpunit/tests/class-test-functions.php	Adds regression tests ensuring cleanup removes both argless and arg-carrying scheduled events.
uninstall.php	Ensures new retry-counter meta key is removed on uninstall.
docs/php-coding-standards.md	Documents the new atmosphere_publish_retry_delays filter in the hook reference list.
.github/changelog/add-publish-retry-backoff	Changelog entry for publish retry/backoff behavior.
.github/changelog/fix-unschedule-all-cron-events	Changelog entry for reliable cron cleanup on disconnect/deactivation.
.agents/skills/code-style/SKILL.md	Updates internal skill reference list to include the new filter.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.