fix(workspace): make scheduled deletion purge reliable

[viethung0o0]

Summary

Scheduled workspace deletion was not completing: the hourly purge crashed while deleting a workspace's messages, so no due workspace was ever removed. This makes the purge correct, resilient, and safe to run alongside live traffic.

What changed

• Purge no longer crashes on message/attachment data. These live in compressed TimescaleDB hypertables that the previous batch-delete could not touch. They are now drained safely by conversation, in small bounded pages, so worker memory stays flat even for very large workspaces.
• One bad workspace can no longer stall the whole purge. Each workspace is torn down in isolation; a failure is logged and retried on the next run while every other due workspace still gets deleted.
• Cleaner teardown. Disconnecting channels no longer throws when an integration's credentials are already gone (Messenger/WhatsApp) and for the built-in web chat channel.
• Deletion timing. The countdown shown to customers is accurate again, and the purge runs on a predictable schedule.
• Heavier work is throttled so it stays gentle on the shared database and is safe under PgBouncer transaction pooling.

Notable

While verifying against TimescaleDB's documentation, the decompression-limit setting used a parameter name that does not exist in TimescaleDB, which would fail at runtime. It has been corrected and centralized in one shared helper (also removing a duplicated statement), with a test that pins the exact name.

Test plan

• pnpm --filter @chatbotx.io/business exec vitest run — 730 passing (adds coverage for the per-workspace guard, hypertable draining, and skip-disconnect-when-no-credentials)
• pnpm --filter @chatbotx.io/database exec vitest run — 378 passing (adds the decompression-parameter-name test)
• pnpm --filter worker test purge-workspaces
• Type-check: database, business, worker-config, worker, webchat, builder
• Lint (Biome) on all changed files
• Verify on staging against a real compressed hypertable: schedule a workspace with historical messages for deletion and confirm the purge completes cleanly

Notes for reviewers

• Message sharding (if enabled) is out of scope here — shard-resident messages are cleaned by the existing per-contact path; a follow-up can extend workspace purge to shards.