Three concurrent refinements from the same work batch:
1. sessiond/manager.rs + runtime_state.rs: switch inhibitor_fds
from HashMap<u64, StdOwnedFd> to HashMap<i32, TrackedInhibitorFd>
so each inhibitor carries both its numeric id and the OwnedFd
that needs to be closed when the caller FD vanishes. The
TrackedInhibitorFd newtype wraps (inhibitor_id, _fd) and lets
reap-on-vanish use the keyed fd handle to take() out of the
map cleanly. The dead_senders code path now collects daemon_fd()
values directly instead of round-tripping through Vec<u64>.
2. wifictl/dbus_nm.rs: introduce AccessPointExport type alias
(OwnedObjectPath, AccessPointInterface) and rename
access_point_interfaces -> access_point_exports returning the
same shape. access_point_paths now derives from the exports
list (avoiding the parse-then-rebuild cycle) and a new
all_access_point_paths() passes through. All call sites in
serve_on_thread get a single coherent exports() call instead
of duplicate path/interface computations.
3. recipes/wip/wayland/qt6-wayland-smoke: correct the relative
symlink target from ../../../local/recipes/wayland/qt6-wayland-smoke
to ../../../../local/recipes/wayland/qt6-wayland-smoke. The
symlink resolves correctly either way (filesystem lookup
succeeds) but the git tree now records a path that is one
level more explicit and matches the canonical Red Bear recipe
symlink convention used elsewhere in recipes/wip/.
Verified via grep that no callers of the old HashMap<u64,
StdOwnedFd> shape remain; all switched to the new TrackedInhibitorFd
type. The sessiond-vs-dead_senders race that motivated the FD
tracking is now correctly closed by taking the daemon_fd handle
out of the map under the same Mutex that updates runtime.inhibitors.
vasilito
9ff09dceaf