…ion metrics

- Fix FIFREEZE/FITHAW ioctl numbers: replace nix::ioctl_write_int! (_IOW)
  with raw libc::ioctl using correct _IOWR constants (0xC0045877/0xC0045878)
- Split export into two phases: only disk flatten runs inside quiesce bracket
  (VM paused ~550ms), checksum+archive run after VM resumes (~6.5s saved)
- Add Instant-based timing metrics to all operations (start, stop, clone,
  export, import, snapshot, quiesce phases)
- Remove debug artifacts (eprintln, sleep) from with_quiesce_async
- Add CLAUDE.md rule #15: no sleep for events

)

* feat: auto-quiesce clone/export on running boxes + REST import_box

Replace require_stopped_for with SIGSTOP/SIGCONT-based PauseGuard for
clone and export operations, allowing them to work on running boxes
without requiring a manual stop. Disable snapshot APIs (to be
re-enabled later). Implement REST client import_box to enable
cross-backend archive portability.

Key changes:
- PauseGuard: RAII guard that freezes VM via SIGSTOP, resumes on drop
- clone_box/export work on running boxes (auto-quiesce via PauseGuard)
- Snapshot APIs return unimplemented error (temporarily disabled)
- REST import_box client: reads archive, POSTs to /boxes/import
- Fix clone_cow Disk RAII leak (.leak() prevents auto-delete)
- Python examples for clone/export/import and local-to-REST migration

* feat: fix FIFREEZE ioctl, minimize VM pause during export, add operation metrics

- Fix FIFREEZE/FITHAW ioctl numbers: replace nix::ioctl_write_int! (_IOW)
  with raw libc::ioctl using correct _IOWR constants (0xC0045877/0xC0045878)
- Split export into two phases: only disk flatten runs inside quiesce bracket
  (VM paused ~550ms), checksum+archive run after VM resumes (~6.5s saved)
- Add Instant-based timing metrics to all operations (start, stop, clone,
  export, import, snapshot, quiesce phases)
- Remove debug artifacts (eprintln, sleep) from with_quiesce_async
- Add CLAUDE.md rule #15: no sleep for events

* refactor: make local_to_rest_migration example self-contained

Start the reference server as a subprocess on a random port
instead of requiring manual server startup. Removes a common
source of confusion and stale-build errors.

…ven detection

ProcessMonitor::wait_for_exit() used a 500ms sleep-based polling loop
(tokio::time::sleep + try_wait), violating the project's Rule boxlite-ai#15:
"No Sleep for Events." This added up to 500ms latency to VM crash
detection during startup (used in guest_connect.rs select! race).

Replace with platform-native event-driven mechanisms:
- Linux: pidfd_open() (kernel 5.3+) + tokio AsyncFd
- macOS: kqueue + EVFILT_PROC + NOTE_EXIT + tokio AsyncFd
- Fallback: 100ms polling for older kernels (< 5.3)

Key design decisions:
- OwnedFd wraps raw FDs immediately after creation (leak-free by construction)
- fcntl O_NONBLOCK checked; falls back to polling on failure
- Block scope for kevent struct (contains *mut c_void, not Send)
- Best-effort race guard via is_alive() after FD setup

Co-Authored-By: Claude Opus 4.6 <[email protected]>