#13 m5 run_in_stdin races early-exit child (EPIPE flake)

m5.rs run_in_stdin does write_all(input).unwrap() on child stdin. When the child exits early (e.g. wiki put ../evil usage check before stdin read), the write races child exit -> BrokenPipe panic (~1/10 under parallel load). Fix in harness: ignore EPIPE on stdin write; exit-code/output assertions still verify behavior.