fix: prevent global lock leak in db_proxy commit() and pool guard

[remo-lab]

Description

This PR fixes a deadlock issue in PandaHarvester caused by a global DB lock not being released when commit() fails.

If a database error occurred during commit(), the application-level lock remained held, which could block all Harvester threads and freeze the system until a restart.

---

Fix

The lock release logic in commit() is moved into a finally block so it is always executed, even when an exception is raised. This follows the same safe pattern already used in rollback().

A small defensive fix is also added in the DB connection pool to ensure a connection is only returned to the pool if it was successfully acquired.

---

Files Changed

• pandaharvester/harvestercore/db_proxy.py
• pandaharvester/harvestercore/db_proxy_pool.py

---

Impact

• Prevents system-wide deadlocks on database failures
• Allows the system to recover gracefully without restarts
• No change to normal success paths

[remo-lab]

Hi @mightqxc , this PR fixes a deadlock caused by a DB lock not being released on commit() failure.
The change ensures the lock is always released and adds a small guard in the connection pool.
Would appreciate a review. Thanks!