run_valgrind_tests.sh treats only exit 100 as failure, so crashes look green

[leo-aa88]

Summary

run_valgrind_tests.sh only fails when valgrind exits 100 (--error-exitcode=100):

if [[ $valgrind_exit_code -eq 100 ]]; then
    echo "Valgrind detected memory issues in $f"
    exit 1
fi

Any other non-zero — missing binary, ASan startup SIGSEGV (typically 139), interpreter crash — is treated as success and the loop continues. That is how #186 stayed hidden: valgrind never ran the program (0 allocs, 0 frees) and the script still exited 0.

Why this matters

Even after #202 (make valgrind builds a non-ASan brainrot-valgrind), a mis-invoked script (./run_valgrind_tests.sh with no args / ASan ./brainrot) or a hard crash still reports green.

Expected Brainrot parse/semantic errors must keep passing: those fixtures exit non-zero on purpose, and valgrind then forwards that exit code when there are no leaks. Do not fail on every non-zero.

Suggested direction

Fail (or at least abort the suite) on unexpected valgrind/child deaths without treating application-level error exits as leaks. For example:

• treat signal exits (128+N, e.g. 139 SIGSEGV / 134 SIGABRT) as failure
• treat valgrind usage errors / missing binary as failure
• keep allowing the interpreter's own non-zero exits when valgrind reports no errors

Spotted while reviewing #202.