docs: correct cancel_timer, autovacuum and fetch_unprocessed docstrings, document ping

[lesnik512]

For 0.15.1, no issue: four docstrings and one docs paragraph that were wrong or left out something operators need. No code change.

• cancel_timer: the docstring said a claimed row's delivery "completes normally". If the worker that claimed it died, the lease stays set until lease_ttl_seconds passes and the row is delivered again, and cancel_timer keeps returning False the whole time. The docstring now says so.
• autovacuum.py: the module docstring said the default vacuum threshold grows with the bloated table size. It grows with the live row count (reltuples), so the real problem is a large backlog, which delays vacuum exactly when churn is highest. The recommendation itself doesn't change.
• fetch_unprocessed: no index covers the queue filter, so a call for a rare or empty queue reads the whole table. The docstring now says to keep it for tests and occasional inspection.
• broker.ping(): it had no docstring. Its database check takes a connection from the engine's pool, so it also fails, after up to 5 seconds, when the application's own sessions hold every pooled connection. The new docstring and the Prometheus/OpenTelemetry setup page, which called make_ping_asgi a liveness probe, now recommend it as a readiness probe, or leaving the pool room for it behind a liveness probe.

That last point is the review finding about ping() failing while the pool is busy. Since #230, the subscribers can't exhaust the pool on their own. The case left is the application's own sessions using every connection, and failing the probe there is arguably right. Making ping() skip the pool would need SQLAlchemy internals, so this PR documents the behaviour instead.

Verification

• 704 tests pass in docker.
• ruff and ty are clean, and the docs build.

[github-actions]

Benchmark gate

✅ gate passed

scenario	msg/s	delete/msg	WALrec/msg	WALB/msg	fpi	upd	del	dead_tup
consumer/w1/b10	1235	1.000	6.69	913	242	5000	5000	10000
consumer/w1/b100	1364	1.000	7.09	1143	243	5000	5000	10000
consumer/w2/b10	1552	1.000	6.69	914	242	5000	5000	10000
consumer/w2/b100	1758	1.000	7.09	1143	243	5000	5000	10000
consumer/w4/b10	1589	1.000	6.72	940	242	5000	5000	10000
consumer/w4/b100	1897	1.000	7.09	1142	243	5000	5000	10000
consumer/w1/b100/tfbs100	6018	0.010	6.07	1121	245	5000	5000	10000
consumer/w1/b100/interval1-10	1403	1.000	7.09	1143	243	5000	5000	10000
producer/w1/b100	2883	0.000	3.04	597	1	0	0	0
producer/w1/b100/ppt1	717	0.000	4.04	637	1	0	0	0

Gated (fails the build): delete_calls + tuple counters (upd/del/ins) + the producer's insert_calls, exact; select_calls and other_calls within +2; wal_records within a 10% band; fetch_calls within 2x. msg/s, WAL bytes and total calls are informational (timing/FPI noise).