Crash respawn never kills old child — can hang forever

[kKaskak]

At server.rs:163-164, the old child handle is dropped without child.kill() or wait(). If the child is hung but not closing stdout, reader threads block on read() forever, the joins block, and the crash notice never fires. Combined with the JobObject issue, this creates a permanently-stuck runtime pinning a CPU core. Need explicit child.kill() before respawn.

[kKaskak]

Verdict: valid with edge cases.

The process handle is never killed or waited, so a runtime that keeps running after its stdio closes can survive while a restart launches another runtime. If the child hangs with stdio open, the current code actually never reaches the crash notice because the reader joins block. So the issue is valid, but the failure modes split into two cases: add explicit child lifecycle control, including kill-on-restart/shutdown and wait/cleanup.