test: read the db clock with scalar_one so ty accepts it under sqlalchemy 2.1

[lesnik512]

Fixes the red lint leg of the scheduled dependency check (run 36128642533).

SQLAlchemy 2.1.0 retyped Result.scalar() from Any to Optional[_T], so ty 0.0.84 reports Any | None + timedelta in test_mark_pending_with_lease_uses_db_clock. SELECT clock_timestamp() always returns exactly one row, so scalar_one() states the real contract and is what the next_at read two lines below already uses.

Test-only; no floor or cap change. Verified: just lint-ci green on sqlalchemy 2.0.54 and 2.1.0; tests/test_integration.py 89 passed against Postgres.

[github-actions]

Benchmark gate

✅ gate passed

scenario	msg/s	delete/msg	WALrec/msg	WALB/msg	fpi	upd	del	dead_tup
consumer/w1/b10	980	1.000	6.79	919	242	5000	5000	10000
consumer/w1/b100	965	1.000	6.83	918	243	5000	5000	10000
consumer/w2/b10	1138	1.000	6.78	950	242	5000	5000	10000
consumer/w2/b100	1333	1.000	6.73	957	243	5000	5000	10000
consumer/w4/b10	1015	1.000	6.82	995	261	5000	5000	10000
consumer/w4/b100	1502	1.000	6.86	1036	244	5000	5000	10000
consumer/w1/b100/tfbs100	5257	0.010	6.07	1116	243	5000	5000	10000
producer/w1/b100	2203	0.000	3.04	584	0	0	0	0

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