db: prefer batched deletes

[DolceTriade]

Our invocation output table is huge and deleting so many rows causes major lock contention. Delete them a few at a time instead to decrease lock contention.

A batch size of 10 was intentionally chosen to deal with large write amplification from cascading deletes.

[Copilot]

Pull request overview

This PR implements batched deletion of old invocations to reduce database lock contention during periodic cleanup operations. Instead of deleting all expired invocations in a single transaction, the cleanup process now deletes them in small batches with delays between batches.

Key changes:

• Added delete_invocations_batch method to the DB trait that accepts a limit parameter for batch size
• Modified the periodic cleanup task to loop through batches of 10 invocations with 100ms delays between batches
• Implemented the batched deletion in both SQLite and Postgres database backends using a SELECT-then-DELETE pattern

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
blade/state/lib.rs	Adds new delete_invocations_batch method to the DB trait interface
blade/main.rs	Replaces single bulk delete with batched deletion loop in periodic cleanup task, adding batch size and delay constants
blade/db/sqlite/mod.rs	Implements delete_invocations_batch for SQLite using SELECT-then-DELETE pattern and adds comprehensive test coverage
blade/db/postgres/mod.rs	Implements delete_invocations_batch for Postgres using SELECT-then-DELETE pattern and adds comprehensive test coverage

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.