fix(db-vacuum): lower default batch_size from 1000 to 200
The orphaned-row queries (NOT EXISTS subqueries) require scanning the
full log/artifact table. At batch_size=1000 on Postgres this takes ~9-11s,
exceeding asyncpg's default command timeout. At 200 the same query
completes in ~1.6s.
Tested against a live Postgres instance with 2.7M logs and 102K flow runs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>