fix: close SQLite connections in delivery_ledger, async_delegation, and verification_evidence ledgers

[JonthanaHanh]

Summary

All three ledger modules (delivery_ledger.py, async_delegation.py, verification_evidence.py) create a new sqlite3.Connection via _connect() on every database call but rely on the context-manager protocol (with conn:) which only handles commit/rollback — it never calls conn.close(). Each call leaks one file descriptor; under sustained gateway or cron load the process hits the OS fd limit.

Root Cause

sqlite3.Connection.__enter__ returns self and __exit__ calls commit() or rollback() but does not close the connection. Every with _connect() as conn: block leaks a connection until GC runs (which is non-deterministic).

Fix

Wrap every _connect() call with contextlib.closing() so the connection is explicitly closed on scope exit:

# Before
with _DB_LOCK, _connect() as conn:
    conn.execute(...)

# After
with _DB_LOCK, closing(_connect()) as conn:
    conn.execute(...)

Files Changed

• gateway/delivery_ledger.py — 5 call sites
• tools/async_delegation.py — 13 call sites
• agent/verification_evidence.py — 3 call sites

Fixes

Fixes #69678
Fixes #69567

[alt-glitch]

This was generated by AI during triage.

Duplicate of #69681: both close the same SQLite connections in delivery_ledger, async_delegation, and verification_evidence. #69681 is the earlier open patch and additionally covers schema-init and failure paths with regression tests.

[teknium1]

Closing: same fix as #69681 (which was submitted earlier and included tests) — the bug class is resolved on main via #70985, which credits both base PRs. Thanks for the minimal, correct patch.