LIFT-2452: Restore async FirehoseClient to fix player disconnections

[mason-chester-rp]

Summary

• Restores fire-and-forget async Firehose behaviour lost during the LIFT-1806 AWS SDK v2 migration
• Swaps FirehoseClient (sync) for FirehoseAsyncClient (async) in FirehoseDb — putRecordBatch() on the async client returns a CompletableFuture that is intentionally not awaited, so Tomcat threads are never blocked on the Firehose HTTP round-trip
• Bumps version to 0.3.6.18

Root Cause

The original SDK v1 code explicitly used AmazonKinesisFirehoseAsync and putRecordBatchAsync(). The LIFT-1806 migration switched to the synchronous FirehoseClient, silently blocking every event POST on the Firehose round-trip. Players have a ~10s read timeout, causing widespread disconnections.

Prod impact confirmed (last 24h, ecs/prod/player/snooze/event-api):

• ~1,886 request very slow: FirehosePostHandler_POST_events warnings
• ~3,871 ClientAbortException: SocketTimeoutException

Test plan

• Build passes (./gradlew test)
• Deploy to dev/e2e and confirm request very slow warnings and SocketTimeoutException errors drop to zero on the event API
• Confirm Firehose stream delivery is unaffected (records still arriving in S3/Redshift)