libkernel: pthread_atfork, mlockall and getdtablesize are absent (XPSemu, VLC); branch ready #3628

libkernel lacks pthread_atfork (U9t5kJAWPnA), mlockall (x7g7Ebeo8-U) and getdtablesize (sZuwaDPATKs). From the relink registry order they are predicted stops of XPSemu (PPSA97358, indices 255, 201, 395) and VLC (PPSA85300, getdtablesize at 354 after #3556); not yet observed in a run. XPSemu calls pthread_atfork from a constructor (util/rcu.c:488-495), so that one must return.

Branch ready: screamm:feat/libkernel-xpsemu-process (three commits on main a11c8f7f: f37bccf6 feat(libkernel): implement pthread_atfork, 87f4aa33 feat(libkernel): add mlockall, 774670de feat(libkernel): add getdtablesize). I cannot open it as a PR because of the intake check; anyone eligible is welcome to carry it, keeping the commits as they are.

• pthread_atfork (new Pthread/Posix/Atfork.cpp): appends the three handlers to a locked list and returns 0 (ENOMEM on allocation failure), as FreeBSD 11.4 libthr thr_fork.c (8c639945); null handlers accepted; the handlers never run because fork throws here, which is recorded, and the list is exposed through PthreadAtforkHandlers_nid_no_patch() for a future fork. New guest_pthread_atfork checks order, null handlers, concurrent registration from four threads, and that nothing runs.
• mlockall(flags): EINVAL for 0 or bits outside MCL_CURRENT|MCL_FUTURE (FreeBSD vm_mmap.c); valid flags throw naming the flag, because no mapping path can lock future mappings and mlock pins host pages range by range without a record to build on; XPSemu only calls it with -overcommit mem-lock=on. guest_mlock gains the cases.
• getdtablesize: FreeBSD's syscall returns min(RLIMIT_NOFILE, kern.maxfilesperproc) (kern_descrip.c), and no public source (shadPS4, KytyPS5, OpenOrbis, fpPS4 placeholder, SharpProspero) gives the console's descriptor limit, so it throws like getrlimit(RLIMIT_NOFILE) still does; the export alone lets VLC and XPSemu load, and both call it only from their own closefrom, which XPSemu does not use. guest_getrlimit checks the throw.

Tested on Windows 11 Pro 26220, WinLibs MinGW-w64 GCC 15.2.0 r7, Ninja, Release, BUILD_TESTING=ON, full build including libs: the three tests 3/3; ctest -j8 -E "^agc_driver_|^video_out_flip" 534/537, the failures being the known windows_exceptions (#3411), guest_kernel_sys (#3510) and guest_filesystem (#3518); with main's Process.cpp the extended getrlimit test fails; tools/check_conventions.py --base main and git diff --check pass. Not run on Linux.

Written with Claude Code assistance (implementation by a subagent, reviewed and validated in this session).