Batch Sakana log writes for higher throughput

[hammadtq]

Context

mem/sakana.py currently calls the Weaviate writer once per log event. Under sustained traffic (~100 logs/s) this creates high HTTP overhead.

Proposal

• Buffer events in memory for N items or ≤ 50 ms. * Use client.batch.add_data_object to send one batch to Weaviate. * Keep the public API await write(event) unchanged.

Acceptance

Criteria - Integration test posts 100 logs → ≥ 3× faster wall time. - Batch size/delay tunable via env vars SAKANA_BATCH_SIZE, SAKANA_BATCH_MS. - All events still appear in class MemoryEvent.

Code reference

See TODO in mem/sakana.py.