v0.4.1

fix: kill code-server process group when codetap exits

When codetap run --stdio exits (relay pipe breaks, SSH drops), the
spawned code-server processes (sh wrapper + node) were orphaned and
reparented to init, leaking memory indefinitely.

- Change ServerRunner.Start to return wait/stop functions instead of blocking
- Use Setpgid to put sh+node in their own process group
- Set Pdeathsig as kernel safety net for unexpected codetap death
- Signal the process group (-pgid) instead of just the direct child
- RunStdio now calls stop() when the relay ends, then drains wait()