A guest signal does not interrupt futex, epoll_wait or a finite select

[Max042004]

A thread parked in these host waits notices a queued signal only when its
own polling quantum or timeout ends. Linux returns EINTR at once.

PR #377 fixed nanosleep not returning on a queued signal; futex,
epoll_wait and select have the same defect. Measured on Apple M4,
against Linux 6.12 under qemu with HVF.

signal to handler, median of 12	elfuse	Linux 6.12
nanosleep (after #377)	0.080 ms	0.044 ms
futex, finite timeout	40.7 ms	0.044 ms
futex, indefinite	40.6 ms	0.044 ms
epoll_wait, indefinite	125 ms	0.047 ms
select, 1 s timeout	779 ms	0.050 ms
epoll_wait, 1 s timeout	777 ms	0.047 ms

futex re-checks signal_pending on a 100 ms quantum
(FUTEX_OS_SYNC_POLL_CAP_NS) and an indefinite epoll_wait on a 200 ms kevent
slice. A finite select or epoll_wait hands the guest's whole timeout to
pselect or kevent with no wakeup pipe fd and no slice, so the signal waits
for the timeout.

Reproducer: a sibling thread parks in each wait, the main thread sends
SIGUSR1 with pthread_kill after a settle delay, and the handler stamps
CLOCK_MONOTONIC.

[xalestar]

I'll start with select and epoll_wait, which can join the wakeup pipe the way poll already does, and take futex after #377 lands.

[jserv]

Utilize frama-c-mcp to introduce formal verification of guest signal handling.