perf: batch path changes via PathsMut on Windows

[sapphi-red]

Implement Watcher::paths_mut for the Windows backend so that bulk add/remove operations cross the server-thread channel as a single Action::StageAndCommit message instead of N Action::Watch/Action::Unwatch messages with N acks. Each staged change still goes through the existing per-path add_watch/remove_watch, so individual semantics and the event filter are unchanged.

Cuts ~27-37% off the windows_paths_mut bench at 100-1000 paths by amortizing the channel round-trip across the batch.

windows_paths_mut/nonrecursive/100
                        time:   [2.1099 ms 2.1233 ms 2.1366 ms]
                        change: [−33.148% −31.854% −30.649%] (p = 0.00 < 0.05)
                        Performance has improved.
windows_paths_mut/recursive/100
                        time:   [2.2734 ms 2.2883 ms 2.3029 ms]
                        change: [−31.250% −29.819% −28.525%] (p = 0.00 < 0.05)
                        Performance has improved.
windows_paths_mut/nonrecursive/500
                        time:   [9.3519 ms 9.3980 ms 9.4457 ms]
                        change: [−38.313% −37.085% −35.954%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild
windows_paths_mut/recursive/500
                        time:   [9.4852 ms 9.5319 ms 9.5815 ms]
                        change: [−39.030% −37.602% −36.251%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  4 (4.00%) high mild
windows_paths_mut/nonrecursive/1000
                        time:   [18.885 ms 19.009 ms 19.141 ms]
                        change: [−37.947% −36.939% −35.987%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 5 outliers among 100 measurements (5.00%)
  5 (5.00%) high mild
windows_paths_mut/recursive/1000
                        time:   [18.889 ms 18.998 ms 19.113 ms]
                        change: [−38.237% −37.280% −36.434%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  7 (7.00%) high mild