fix(backend): address database connection issues and pool exhaustion (#5496)

[MekonMAC]

This PR addresses critical database connection issues under high load as described in #5496.

Changes:

1. Fixed Deferred Loader Failure: Added session.refresh(deduplication_event) in create_deduplication_event to ensure the id is populated before it's accessed for logging.
2. Prevented 'Idle in Transaction': Added a SQLAlchemy checkout listener to perform a rollback() on connection checkout. This ensures connections are returned to the pool in a clean state and prevents sessions from being stuck in 'idle in transaction' state, which is a major cause of pool exhaustion.
3. Optimized SQLite Performance: Enabled WAL (Write-Ahead Logging) and Normal synchronous mode for SQLite to improve concurrency and performance in local/dev environments.

These changes significantly improve the reliability of the connection pool both with direct PostgreSQL connections and when using PgBouncer.

Fixes #5496

[greptile-apps]

Target branch is not in the allowed branches list.

[vercel]

Someone is attempting to deploy a commit to the KeepHQ Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[MekonMAC]

Hi Team! It seems that the CI for this PR is failing due to some global issues in the main branch:

1. Release JSON Schema: Failing with npm error Missing: @types/react (mismatch between package.json and lock file).
2. Unit Tests: Failing with RecursionError: maximum recursion depth exceeded in keep/api/api.py.

Could you please take a look at these infrastructure/main-branch issues? Once fixed, I'll rebase my PR. Thanks! 🚀