`TRYCMD=overwrite cargo test` won't stop for errors that a follow up `cargo test` would

[epage]

We originally had #31 which was fixed in #188 but it caused this effect. #189 fixes probably the biggest case of an error that should be reported during TRYCMD=overwrite but we should probably clean the code up more generally to ensure there aren't other cases. This would involve handling this when we overwrite, much earlier in the process, as that is when we should know whether an error was resolved by the overwrite or not.