Every test that passes the --semantic phase can also serve as a --compile or --compile-firm test where the runtime behavior is unspecified.

Specifically, these tests may have undefined behavior, so it doesn't even make sense to run the resulting executable. Yet, they might provoke crashes in the middle- or back-end, so it could still be worthwhile to test them.

Not sure what the resulting mode should be called; compile is already taken.

This subsumes #1 (closed), which proposes an entire new test category for this kind of testing. This proposal allows reuse of all semantic tests for this test category.