refactor: one seconds-interval helper; coverage tracks greenlet switches

[lesnik512]

Cosmetic, no issue. The three make_interval(0, 0, 0, 0, 0, 0, :param) calls in client.py now go through one _seconds(name) helper:

• the fetch's lease cutoff;
• the INSERT's activate_in;
• mark_pending_with_lease's retry delay.

Two comments that explained the six zeros and the clock-skew reason are folded into its docstring.

Why not make_interval(secs => ...)

That was the original idea. SQLAlchemy's func has no named-argument form, so it would need a custom compiled element. It would also change the SQL text, which moves the pg_stat_statements query ids the benchmark gate counts. The helper keeps the SQL byte for byte.

Verification

• The compiled fetch statement and all eight INSERT variants are identical to main's.
• just bench-check passes, which also covers mark_pending_with_lease.
• After rebasing onto main, 762 tests pass in docker at 100% coverage, and ruff and ty are clean.

Also in this PR: coverage tracks greenlet switches

This PR's pytest (3.11) job failed twice in a row, though all 744 tests passed both times. The coverage gate reported tests/test_integration.py:1512 as never run. That line is a publish loop in test_batched_flush_dlq_row_stays_inline, and the PR doesn't touch it.

Main has the same flake. I ran that test 15 times under coverage and counted the runs that reported line 1512 missed:

	missed
main, 3.11	5 of 15
this branch, 3.11	7 of 15
this branch with concurrency = ["greenlet", "thread"], on 3.11, 3.12 and 3.13	0 of 15 each

Cause: SQLAlchemy's async layer runs every statement in a greenlet. Coverage's C tracer loses its place across a greenlet switch it isn't tracking, so a line that runs after await session.execute(...) can go unrecorded. #235 worked around this with core = "sysmon", but Python 3.11 has no sys.monitoring, so the 3.11 job kept the C tracer and the flake.

The change: [tool.coverage.run] now sets concurrency = ["greenlet", "thread"] and drops core = "sysmon" and its no-sysmon warning filter.

• sysmon goes away: coverage 7.16 can't combine sysmon with greenlet tracking. It falls back to the C tracer with a no-sysmon warning, and the old filter would have hidden that. Every version now uses the same tracer.
• Speed: the full suite runs in about the same time, 36 s.
• Branch coverage: just test-branch misses 16 partial branches against main's 20. The tracer was dropping those 4 too.

[github-actions]

Benchmark gate

✅ gate passed

scenario	msg/s	delete/msg	WALrec/msg	WALB/msg	fpi	upd	del	dead_tup
consumer/w1/b10	1935	1.000	6.71	920	242	5000	5000	10000
consumer/w1/b100	1676	1.000	7.11	1154	243	5000	5000	10000
consumer/w2/b10	2476	1.000	6.72	924	242	5000	5000	10000
consumer/w2/b100	3081	1.000	7.11	1154	243	5000	5000	10000
consumer/w4/b10	2498	1.000	6.73	940	242	5000	5000	10000
consumer/w4/b100	3104	1.000	7.11	1153	243	5000	5000	10000
consumer/w1/b100/tfbs100	9976	0.010	6.11	1130	243	5000	5000	10000
consumer/w1/b100/interval1-10	2030	1.000	7.11	1154	243	5000	5000	10000
producer/w1/b100	5741	0.000	3.04	597	1	0	0	0
producer/w1/b100/ppt1	1738	0.000	4.04	635	0	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).