fix: logic to handle missing ref files was throwing if we see mixed statuses

Observed in this nightly

BrunelSys test has two refs, one for -opt and one for -dbg. Tests were only failing on the -dbg platform.

The check for handling the case where we were missing references simply checked if all platforms were failing with a dict["Causes"] == "unexpected missing reference file".
But at the point in the script where this check was performed, it is totally valid for some tests to have passed and thus not have the field Causes in their dict.

Previously not observed because the case is pretty rare, we mostly either have all tests pass and thus don't get to the problematic line or have all tests fail, thus all tests have the needed Causes field.

Fixed by moving the handling of missing references a bit lower in the code where I can actually profit more from all the checks that are done for the default scenario anyways.