fix-inventory-sync-race-condition

The sync service reads current stock before processing all events, so a second event can replay against an outdated count.

I added per-SKU locking and idempotency checks so duplicate events are skipped instead of replayed.

The race spec dispatches two events for the same SKU and verifies that only one stock write wins the lock.

Lock coverage confirms that completed event ids are cached, while new event ids can still update stock normally.

The build failed on an API timeout while preparing the integration fixture. The code changes are present, but this run needs a retry.

I stopped after the timeout so the failure state stays explicit instead of hiding it behind another automatic retry.