limactl start fails with stale ha.sock/ha.pid from unclean shutdown

If the system shuts down before `limactl start --foreground` finishes cleanup (signal handler in `hostagent.go`), stale `ha.pid` and `ha.sock` files remain in the instance directory. On the next boot, `limactl start` tries to connect to the stale `ha.sock`, gets connection refused, and exits with a fatal error instead of cleaning up and starting fresh.

The hostagent command already has `defer os.RemoveAll(pidfile)` and signal handling via `signalCh`, but macOS may kill the process before cleanup completes during a fast shutdown. `limactl start` could check whether the PID in `ha.pid` is actually running before attempting to connect to `ha.sock`, and clean up stale files if it's not.

This is one of several issues that make Lima unusable for unattended auto-start via LaunchDaemon. See also #2252.

**Environment:** macOS 15.7.7, Lima 2.1.x (fork), LaunchDaemon with `RunAtLoad`.

Start at the limactl start path and the signal handling in hostagent.go. Trace how ha.pid and ha.sock are checked before connecting, and inspect the existing cleanup behavior. Done means a stale PID/socket from an unclean shutdown is removed and the instance starts normally; verify with the relevant Lima start and hostagent tests if available.

Written by the indexing model from the issue text.