Fix usage of reentrant functions in ext/posix

[arnaud-lb]

• It's not necessarily an error when sysconf(_SC_GETPW_R_SIZE_MAX) returns -1, as specified by posix (and the musl implementation always returns -1). Pick an initial buffer size in this case.
• Reentrant variants return an error number an may not set errno
• Implement retry logic for ttyname_r()
• Fix retry logic for getpwnam_r() (pw would be NULL after the first try)
• Test retry logic by setting the initial buffer size to 1 in debug builds

Tests for these functions would fail under musl due to sysconf(_SC_GETPW_R_SIZE_MAX) returning -1 (which is allowed by the standard). The test suite passed on other platforms, but these functions could fail under unusual conditions. Setting buflen to 1 ensures that the retry logic is tested.

[devnexen]

I think a somewhat similar PR was started earlier but maybe I confuse.

[arnaud-lb]

Indeed, there appears to be some overlap with #13922

[devnexen]

Looking good. Do you really want to keep the debug part ? :)

[arnaud-lb]

Thank you!

Yes, this ensures that the retry code paths are tested. This doesn't negatively impact the duration of the test suite.

[iluuu1994]

Is this responsible for the failures in nightly? https://github.com/php/php-src/actions/runs/8682420096/job/23806945143

[arnaud-lb]

It's possible. I will check!

[dstogov]

The nightly failures are definitely related to this patch.
See https://github.com/php/php-src/actions/runs/8655545925/job/23734685655
ext/posix/tests/posix_getpwnam_basic_01.phpt and ext/posix/tests/posix_getpwuid_basic.phpt are failed only on MAC and only with DEBUG_ZTS build. It seems getgrnam_r() "test retry logic" doesn't work properly.