feat: add self-healing sync queue for failed writes

[faisalahammad]

Summary

When Elasticsearch/OpenSearch is briefly unreachable, write operations (index, delete, bulk index) fail silently. The WordPress database and search index drift out of sync until the next full re-index. This adds a persistent journal of failed writes and a WP-CLI command to replay them once the search server is back.

Fixes #4226

Changes

New: FailedWrites.php

Journal class owning a {prefix}ep_failed_writes table created via dbDelta. Captures failed index/delete operations with dedup by (indexable_slug, object_id). Methods: record(), get_pending(), delete_entries(), count_pending().

elasticpress.php

Bump EP_VERSION to 5.4.0. Instantiate FailedWrites::factory() on plugin load.

Command.php

Add replay-queue subcommand with --limit, --dry-run, --skip-health-check flags. Replays each journal entry by re-indexing or re-deleting, clears on success, records new error on failure.

Before: No way to recover from transient failures without a full re-index.
After: wp elasticpress replay-queue replays captured failures.

Why: Matches the reporter's request for a WP-CLI-based recovery mechanism and avoids the cost of a full re-index.

Indexable.php

Fire ep_after_delete_{slug} action after delete_document() so delete failures are capturable. Index and bulk-index paths already had hooks.

Upgrades.php

5.4.0 upgrade routine creates the journal table via dbDelta on first install and on upgrade.

uninstall.php

Drop the ep_failed_writes table on uninstall. Multisite: each site's table is dropped in the per-site loop.

New: TestFailedWrites.php

13 tests (30 assertions): record inserts, dedup by slug+id, get_pending ordering, delete_entries, count_pending, delete/update failure capture, dry-run CLI, live replay CLI, multisite blog_id capture, table creation on upgrade.

Testing

Test 1: Journal captures failed writes

1. Point ep_host to an unreachable host
2. Create or update a post
3. wp db query "SELECT * FROM wp_ep_failed_writes" shows a row
4. Updating the same post again still shows one row (dedup works)

Test 2: Replay restores sync

1. Restore reachable ep_host
2. wp elasticpress replay-queue shows success with progress bar
3. Journal table is empty after replay
4. Document is searchable in ES

Test 3: Dry-run mode

1. wp elasticpress replay-queue --dry-run shows entries without executing

Test 4: Unit tests

1. vendor/bin/phpunit --filter TestFailedWrites - OK (13 tests, 30 assertions)
2. Full suite: 5 pre-existing failures in TestDocuments and TestWooCommerceProduct (reproduce on clean develop, not a regression)

Test 5: Uninstall

1. wp plugin uninstall elasticpress drops the journal table