fix(luggage): serialize shim-writing validate tests to close fork/exec race

[joshjhall]

Summary

• Mark the three shim-writing tests in crates/luggage/src/installer/validate.rs with #[serial_test::serial] to close the same fork+ETXTBSY race that fix(luggage): serialize shim-writing idempotency tests to close fork/exec race #459 fixed for installer::idempotency.
• The three tests (matching_output_passes, mismatched_output_returns_validation_failed, propagates_env_to_subprocess) use the identical write_shim + Command::output() pattern but were never serialized.

Root cause

CI run 25982111224 — the merge-to-main of #479 — failed on Ubuntu only with:

test installer::validate::tests::mismatched_output_returns_validation_failed ... FAILED
thread '...' panicked at crates/luggage/src/installer/validate.rs:127:17:
assertion failed: message.contains(\"1.95.0\")

The assertion fires inside the `ValidationFailed { message, .. }` arm. The only `message` formats that don't contain the requested version are the launch-failure branches (`binary not found`, `failed to launch`, `exited N: ...`). Under parallel `cargo test`, one thread's in-flight write-FD on the shim raises `i_writecount` while another thread forks for `execve`, and the child returns `ETXTBSY` (rust-lang/rust#100904). Ubuntu's faster fork/exec exposes it; macOS / Windows runners did not reproduce.

This is the same diagnosis and fix recipe that #459 applied to `idempotency.rs` — the `serial_test` dev-dep is already wired in.

Test plan

• `cargo test -p luggage --lib installer::validate` — 4/4 pass
• `cargo test -p luggage --lib` — 151/151 pass
• `cargo clippy -p luggage --tests -- -D warnings` — clean
• Stress: ran `cargo test -p luggage --lib installer::validate` 25× consecutively — all green

🤖 Generated with Claude Code