reactor: the wait's timeout replaces the 250 ms timer, and wakes for a due QUIC deadline

[MDA2AV]

An idle reactor sleeps in io_uring_enter until a completion arrives, and the only one it can count on is the 250 ms ticker. So a QUIC retransmit, ACK or pacing step due in a few ms waited for the next tick. #267 made the deadline correct; nothing woke the loop for it.

This makes the wait itself the timer: it is bounded by the next ticker run (250 ms) or the earliest engine deadline, whichever is sooner. The tickers run from the loop once their interval is up, checked every pass, because a busy reactor's wait never runs out. The IORING_OP_TIMEOUT that used to drive them, its KindTimer completion and its native timespec are gone.

Chosen over #271, which kept the timer and only bounded the wait for QUIC.

Tests

• quic/timer: an idle reactor fires an engine deadline when it is due, not at the next tick: the answer is lost and the peer stays silent, so the retransmit timer fires again and again; each firing is timed against the engine's own ns deadline.
  • main: up to 226-237 ms late ([146, 233, 22, 44, 89], [144, 226, 13, 26, 52] ms)
  • this PR: under 1.1 ms
• quic/timer: a reactor whose last connection is gone does not keep waking for it: nothing resets the tracked deadline when the last connection leaves, so the wait ignores it then. Without that guard the empty reactor woke 996 times/s.
• tcp/read: a quiet connection is reaped while another keeps the reactor busy: the sweeps must not wait for an idle moment. With the tickers run only when the wait times out, the quiet connection was never reaped (644,301 exchanges on the other one over 4 s).
• quic/timer: a deadline that is already due is reported as due… needed the loop to notice a deadline 2 ms late, which it no longer does; any amount past now counts (the clock only moves forward).

All suites pass, pending lists unchanged from main: E2E 234, Unit 60, Chaos 47, Http 44, Tls 151 (the 7 kTLS tests skip without sudo), File 4.

Bench

Interleaved: 5 rounds of 10 s per sample in rotating order, 2 reactors pinned to the P-cores. control is a second build of main, so its delta is the noise floor. Medians against main, req/s · CPU per request:

sample	main	this PR	#271	control (main again)
Tcp/Raw (h1)	883,722/s · 2.285 µs	+4.3% · −4.0%	+1.3% · −1.3%	+3.7% · −3.5%
Http3/Nghttp3Buffered (h3, 16 conns)	782,765/s · 2.552 µs	+1.3% · −1.3%	+0.9% · −0.8%	−0.7% · +0.7%
Tls/OpenSsl (h1s)	365,622/s · 5.533 µs	+0.7% · −0.7%	+0.1% · −0.1%	+2.3% · −2.2%

Every delta is inside the control's own spread, so neither design costs anything measurable under load. Tcp/Raw rerun over 10 rounds, within-round median vs main: this PR +0.3%, #271 −0.8%, control −0.9% - the +4.3% above was noise. Idle reactors wake 4.0-4.2 times/s on main and both PRs alike (the 250 ms tick), for TCP and for QUIC after a 1.35M-request h3 burst has ended.

Size: src +87 −72 in 11 files.