libkernel: fcntl record locks throw on Windows (ProsperoRadio, ProsperoTV after #3649); fix branch ready #4479

With #3649 applied, ProsperoRadio (PPSA99001) stops on #3840's recorded Windows limit: what(): fcntl: record locks are not implemented on Windows, command=12, fd=3 (F_SETLK, at 33.7 s in download0/prosperoradio.log); with the POLLPRI fix (#4478) applied as well, ProsperoTV (PPSA99003) reaches the same line.

Fix branch ready: screamm:feat/libkernel-record-locks-windows: b7cc2856 on top of #3649's two commits on main 33ae282e (depends on #3649 only for the title evidence; the commit itself touches DescriptorControl.cpp and the fcntl test). I cannot open it as a PR under the intake rule while #4081 is open; anyone eligible is welcome to carry it, keeping the commit as is.

• F_GETLK, F_SETLK, F_SETLKW and the old F_OGETLK/F_OSETLK/F_OSETLKW forms now work on Windows with FreeBSD 11.4's checks in FreeBSD's order (kern_fcntl kern_descrip.c:607-752, lf_advlockasync kern_lockf.c:433-487, vop_stdadvlock, 8c639945): SEEK_CUR resolved against the current offset (EOVERFLOW past the maximum, a directory using its emulated cursor), EBADF for a read lock on a descriptor without read access or a write lock without write access (F_GETLK checks the type but not access), EINVAL for a bad type, negative start, unknown whence or a negative length reaching below 0, EOVERFLOW for SEEK_END past the maximum; F_GETLK writes back F_UNLCK with the adjusted start (24 bytes for the old form). Two cases still throw: a SEEK_END lock on a directory (size unknown) and locks on character devices. Linux is unchanged.
• No lock table and no LockFileEx backing, deliberately: FreeBSD's lf_blocks lets only locks of different owners conflict, and every lock belongs to the process, so inside the single guest process F_SETLK never returns EAGAIN, F_SETLKW never waits and F_GETLK always reports F_UNLCK; merge/split and release on close are unobservable. Windows byte-range locks are mandatory and would block the guest's own other descriptors (a shared lock blocks writes even through the locking handle), which breaks the POSIX semantics, so other host processes neither see nor are seen by these locks; recorded in the fcntl TechnicalDebt bullet. A table is a contained addition if multi-process support ever arrives.
• guest_kernel_fcntl: the old Windows "throws" expectation becomes the Linux round trip, plus a new RecordLocksBelongToTheProcess section (round trips, overlap merge and split sequences, two descriptors on one range, a second thread's F_SETLK/F_SETLKW/F_GETLK without conflict, SEEK_CUR/SEEK_END, negative lengths, the 24-byte struct, EINVAL, EBADF on read-only and write-only descriptors, relock after close, EBADF after close); the EOVERFLOW, error-ordering, F_GETLK with F_UNLCK and directory cases are Windows-only, pending Linux CI for the shared part.

Tested on Windows 11 Pro 26220, WinLibs MinGW-w64 GCC 15.2.0 r7, Ninja, Release, BUILD_TESTING=ON, full build including libs: guest_kernel_fcntl, guest_poll, guest_dup, guest_file_vector 3/3; ctest -j8 -E "^agc_driver_|^video_out_flip" 620/620; conventions and git diff --check pass. ProsperoRadio relinked against this build (run 4 runtime, 60 s) opens /download0/radio-browser.sqlite3, starts its service and reaches its HTTP code, then stops at 16.5 s on what(): poll: POLLPRI is not supported by WSAPoll (#4478's fix). Not run on Linux.

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