Cover the EINTR retry in natsSock_WaitReady with two signal arms

[groeneai]

natsSock_WaitReady retries poll() on EINTR since b88704df, but test_natsWaitReady only ever waits on a quiet thread, so no arm in the suite reaches that retry. This adds two arms that deliver a signal to the waiting thread. No library code changes.

Asked for in ClickHouse/ClickHouse#120202 (comment).

Both arms are #ifndef _WIN32:

• no deadline, signals delivered: the mock server makes the socket readable after ~500 ms, so the wait has to return NATS_OK then, not on the first signal.
• deadline timeout, signals delivered: a 50 ms deadline over the same storm still has to end in NATS_TIMEOUT within 40 to 100 ms.

Both also assert that at least one signal was delivered, so neither can pass by not being about signals.

Measured on this branch (Linux, -DBUILD_TESTING=ON -DNATS_BUILD_STREAMING=OFF, testsuite natsWaitReady):

library	first arm	second arm	runs
b88704df, this base	PASSED	PASSED	3/3
cb4edba, poll() unretried	FAILED: Error: 3 - IO Error - (unix/sock.c:56): poll error: 4	not reached	3/3
b88704df with the deadline recompute hoisted out of the retry loop	PASSED	FAILED	2/2

Row 2 is the failure ClickHouse CI reports byte for byte (Cannot connect to Nats last error: (unix/sock.c:56): poll error: 4). Row 3 is why the second arm is not redundant: a retry that re-arms the full deadline instead of what is left of it waits until the storm stops rather than ending on its own 50 ms deadline.

SIGALRM comes from pthread_kill() on a helper thread rather than from setitimer(), because ITIMER_REAL signals the process: the kernel then picks any thread that does not block it, which also cuts short the mock server thread's nats_Sleep(). Targeted delivery keeps the bounds deterministic. The handler stays installed after the storm on purpose, since restoring SIG_DFL would let a signal still in flight terminate the process.

test_natsWaitReady is already in test/list_test.txt, so ctest -R natsWaitReady picks the arms up with no registration change.

#8 carried this test together with its own version of the retry. The retry landed as b88704df instead, so I closed that one; this is the test alone, on top of it.

[rschu1ze]

Upstreamed here: nats-io#1030