fix: erroneous QueryCompiler retry when nothing unaccounted for

[nbauernfeind]

Fixes #6216.

I discovered how to reproduce this by accident (my script was mid-write). Using a debugger I found that the retry list was empty, meaning that all of the non-failing items were successfully written to disk.

[cpwright]

Is there a reasonable way to write a unit test for this?

[nbauernfeind]

Is there a reasonable way to write a unit test for this?

It seems like it might be jdk specific. TestQueryCompiler#testMultiCompileWithFailure simulates the scenario where this is reproduced. When I originally implemented parallel compilation, and it seems is true for CI as well, when there was one failure then all compilation units failed. That required a second loop through the compiler to get the passing ones to write out their class files.

When I reproduced on Friday, it tuned out that all of the non-failing units were properly written to disk causing this retry list to be empty; so I'm not certain how to reproduce in the CI environment.

[nbauernfeind]

Actually, it does not seem to be jvm dependent. It is dependent on the ordering of the requests. Unit test incoming.

[nbauernfeind]

Apparently, the type of error matters; some errors appear to not be able to continue whereas other errors may be overlooked. I've updated the unit tests where the new one fails before the bugfix.