perf(sqlite): add indexes for hot query paths

[benvinegar]

Summary

SQLite-backed workspaces currently scan and sort entire tables for several common reads. This adds six targeted indexes for:

• posts within a session, ordered by creation time
• recently updated posts
• incremental session comments and post threads
• comment lookup and deletion by ID
• assets belonging to a session

The indexes are created with IF NOT EXISTS after legacy schema migrations, so existing Durable Objects and local databases upgrade safely on first startup.

Performance

Synthetic in-memory SQLite benchmark with 100,000 posts, 200,000 comments, and 50,000 assets; values are medians from 30 warm runs:

Query	Before	After	Speedup
Session posts	2.657 ms	0.661 ms	4.0×
Recent posts	16.292 ms	0.015 ms	1,061×
Post counts by session	11.987 ms	2.343 ms	5.1×
Session comments	2.609 ms	0.693 ms	3.8×
Post comments	2.019 ms	0.008 ms	259×
Comment by ID	3.846 ms	0.002 ms	2,284×
Session assets	1.395 ms	0.450 ms	3.1×

Building all six indexes took 124 ms in this synthetic in-memory database. Existing databases pay this construction cost once after upgrading.

Testing

• Added coverage for upgrading an existing unindexed workspace and repeated initialization.
• Added query-plan assertions for each optimized production query.
• Passed unit tests, coverage, typechecks, lint, formatting, Worker integration, security audit, Chromium/WebKit E2E, CodeQL, and Socket checks.
• Independent review found no blockers.

This PR description was generated by Pi using GPT-5.6 Sol