feat(clickhouse): add opt-in async cleanup via setAsyncCleanup()

[lohanidamodar]

Summary

Add setAsyncCleanup(bool) to the ClickHouse adapter. When enabled, cleanup() appends SETTINGS lightweight_deletes_sync = 0 so the HTTP call returns once the mutation is queued, instead of blocking until ClickHouse finishes the DELETE. Default behavior is unchanged.

Why

In production, the maintenance worker invokes cleanup() per project against a shared multi-tenant audit table. The synchronous DELETE routinely exceeded the 30s client timeout:

Failed to cleanup ClickHouse audit logs for project <id>:
ClickHouse query execution failed: Operation timed out after 30026 milliseconds with 0 bytes received

Two factors compound:

1. tenant is not in the table's sort key or skip-index set, so the per-tenant predicate forces a row-level scan inside surviving parts.
2. SharedMergeTree serializes mutations per table through Keeper, so N per-project mutations queue up and the HTTP client times out long before they drain.

The maintenance worker only needs to schedule cleanup, not wait for it.

Why opt-in (not default)

The first iteration of this PR made async unconditional and broke testCleanup, testFind, testCount — they assert on row counts immediately after cleanup(). Making it opt-in via a setter:

• preserves the existing contract for every other consumer (and for the test suite),
• matches the existing setSharedTables / setTenant configuration style on this adapter,
• lets cloud's deletes worker opt in with a single line: $auditAdapter->setAsyncCleanup(true).

Test plan

• composer lint (Pint)
• composer check (PHPStan level max, no errors)
• CI: existing test suite passes (default-sync behavior is untouched)
• Cloud follow-up: deletes worker calls setAsyncCleanup(true) and verifies the 30s timeout is gone in staging

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

This PR makes cleanup() non-blocking on the ClickHouse adapter by appending SETTINGS lightweight_deletes_sync = 0 to the lightweight DELETE statement when the new asyncCleanup flag is enabled. The default remains synchronous (false), so existing behavior and tests are unaffected.

• Adds $asyncCleanup property (default false) with setAsyncCleanup/isAsyncCleanup accessors following the same pattern as the existing sharedTables configuration.
• When enabled, the query-level SETTINGS lightweight_deletes_sync = 0 is appended after the WHERE clause, which is the correct position in ClickHouse's DELETE syntax.

Confidence Score: 5/5

Safe to merge — the change is purely opt-in and the default synchronous path is entirely unchanged.

The asyncCleanup flag defaults to false, leaving all existing tests and callers on the synchronous code path. The SETTINGS lightweight_deletes_sync = 0 clause is placed correctly at the end of the DELETE statement after the WHERE clause, which matches ClickHouse's lightweight-delete syntax. The accessor API follows the same pattern already used for sharedTables. No logic, no data integrity concern, and no test breakage on the current code paths.

No files require special attention. The only changed file is src/Audit/Adapter/ClickHouse.php and the modification is minimal and well-scoped.

Important Files Changed

Filename	Overview
src/Audit/Adapter/ClickHouse.php	Adds opt-in async cleanup via setAsyncCleanup(true); appends SETTINGS lightweight_deletes_sync = 0 correctly after the WHERE clause in the DELETE statement. Default is synchronous, so existing behavior is unchanged.

_{Reviews (2): Last reviewed commit: "feat(clickhouse): add opt-in async clean..." | Re-trigger Greptile}