Wake select and epoll_wait on a queued signal

[xalestar]

Part of #378; futex is left for after #383.

A thread parked in a host wait learns of a guest signal only through the wakeup pipe. pselect6 joined the pipe for an indefinite wait only, and epoll_pwait never did, so the handler ran only once the current 200 ms slice ended.

pselect6 now joins the pipe for every wait except a zero timeout, which does not park and would only drain a wake meant for a sibling. epoll_pwait parks in select() on the kqueue fd plus the pipe, then collects with a zero-timeout kevent(), so the pipe is never registered on the guest's epoll instance. select() rather than poll(), because poll() answers POLLNVAL for a kqueue fd and would never park. A guest with a full fd table puts the kqueue past FD_SETSIZE, so poll.c builds with _DARWIN_UNLIMITED_SELECT and that case parks on a heap bitmap.

Reproduction: tests/test-wait-signal-latency.c parks a sibling thread in each wait on an empty pipe, sends SIGUSR1 with pthread_kill, and takes the median signal-to-handler gap over 8 rounds against a 20 ms bound. It also reads the waiter's CLOCK_THREAD_CPUTIME_ID share over the wait against a 5% bound, since a spinning wait answers the signal at once too. A fourth case fills the fd table before epoll_wait. On 0a63368 every case fails at about 42 ms. With this change all four pass under 1% CPU. A build with the select() park forced to a zero timeout reads 10% and fails; one that skips the park past FD_SETSIZE fails the full-table case at 41 ms.

Validation, rebased on b05f639:

• make check: exit 0. One earlier run failed test-shim-futex-toctou with -ENOSYS in its spin phase, the rt_sigreturn overwrites the restored X8, so a signal taken on an svc re-enters it as syscall 2 #379 symptom that Restore the guest X8 when the shim drops its frame #383 fixes; it passed 5/5 on rerun.
• test-wait-signal-latency, test-wait-process-signal, test-wait-sigmask-signal, test-poll, and all nine test-epoll*: pass
• make test-matrix-elfuse-aarch64 (on the first revision): 291 passed, 1 failed, 10 skipped. The failure is test-sigio (no SIGURG on a TCP OOB byte). It also fails 3/3 on 0a63368 and does not touch select or epoll.
• The qemu lane was not run.

[jserv]

epoll_wait no longer sleeps: macOS poll() answers POLLNVAL for a kqueue descriptor, so the park added in epoll_wait_slice returns at once and an idle epoll_wait(-1) holds a full core. The latency test cannot see this, because a spinning loop also rechecks signals immediately. Details and a select()-based fix are on epoll_wait_slice.

[cubic-dev-ai]

All reported issues were addressed across 2 files (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>

[jserv]

Thank @xalestar for contributing!