fix(skillfs): prevent FUSE test mount leaks

[kongche-jbw]

Description

The tests terminate the foreground mount process with child.kill() (SIGKILL, which SkillFS cannot catch), so the FUSE mount is never unmounted and leaks under target/tmp.

This PR:

• Adds stop_mount_child: sends SIGTERM so SkillFS takes its normal unmount path, waits a bounded time for graceful exit, then force-unmounts as a fallback (fusermount3 -u / -z / umount -l) and finally SIGKILLs to reap the child. Replaces all 4 child.kill() call sites; run_briefly now takes the mountpoint so teardown always targets the mounted path.
• Adds a bounded, non-panicking, non-blocking force_unmount_path on MountHandle, invoked from both Drop and the unmount() error path so every teardown route is covered.

Related Issue

closes #1328

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• CI/CD or build changes

Scope

• skillfs (SkillFS)

Checklist

• My code follows the project's code style
• I have added tests that prove my fix is effective (reuses existing tests; the target/tmp mount count no longer grows after a run)
• For skillfs: cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, and cargo test --workspace pass
• Lock files are up to date

Testing

• Verified with the CI-pinned 1.86.0 toolchain: fmt --check / clippy -D warnings / build all clean.
• cargo +1.86.0 test --test cli_startup_tests — 47 passed; grep -c skillfs /proc/mounts is unchanged before/after (it would increase by 1 before the fix).
• Re-ran the previously-leaking non_tmp_transport_paths_pass_gate in isolation — no new target/tmp mount appears.

Additional Notes

MountHandle::Drop does not help the SIGKILL case directly (a SIGKILLed process never runs Drop); it is a defensive safeguard for library consumers and in-crate tests. The actual fix for the leak is the SIGTERM change on the test side.