truncate on /dev/shm drops the guest's record locks #469

truncate() on a /dev/shm path drops every POSIX record lock the guest holds on that file. sys_truncate in src/syscall/fs.c reaches a /dev/shm leaf through shm_open_leaf, truncates through that descriptor and closes it. On macOS, closing any descriptor for a file releases all of the process's fcntl record locks on it, and guest F_SETLK locks are host record locks.

Measured with a static aarch64 probe: open /dev/shm/elfuse-lock-probe, take an F_SETLK write lock, call truncate(path, 100), then have a forked child check the lock with F_GETLK.

Linux's vfs_truncate opens nothing, so the lock survives. #463 removed the same open-and-close from the regular-file path of truncate after review; the /dev/shm branch was left as it was.

Goal: truncate on a /dev/shm path leaves the guest's record locks in place. The branch opens the leaf because truncate(2) has no nofollow form, so a fix would have to keep that protection without closing a descriptor on the file.