codex/fix sqlite workspace write contention

[drew]

Summary

Let concurrent SQLite gateway writes wait through short lock contention instead of failing workspace creation.

Related Issue

Stability investigation stacked on #3607 and ultimately #3583. Kubernetes managed-workspace run 35856895823 failed two parallel creates after SQLite returned database is locked at about five seconds.

Changes

• Set a bounded 30-second busy timeout for SQLite store connections.
• Add a contention test that holds a write lock for six seconds and verifies a concurrent workspace insert succeeds after release.
• Document the gateway's SQLite lock wait.

Testing

• Focused SQLite contention test: passed.
• mise run pre-commit: passed.
• mise run ci: passed.
• Expanded GitHub E2E on stack tip: pending.

Checklist

• Signed-off conventional commit
• Architecture documentation updated
• Full local CI passed
• Kubernetes managed-workspace E2E passed

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[drew]

Superseded by clean stacked PR #3622 in GitHub Stack #3623. The selected fix is preserved there without the intervening campaign changes.