fix counting of failures in PyTorch tests

[Flamefire]

(created using eb --new-pr)

The RegExp handling was wrong, e.g. for an output of:

SKIPPED [2] distributions/test_constraints.py:83: `biject_to` not implemented.
FAILED distributions/test_constraints.py::test_constraint[True-constraint_fn5-False-value5]
FAILED distributions/test_constraints.py::test_constraint[True-constraint_fn7-True-value7]
============= 2 failed, 128 passed, 2 skipped, 2 warnings in 8.66s =============
distributions/test_constraints failed!

the summary_matches_pattern2 will yield "2 failed, 128 passed, 2 skipped, 2 warnings in 8.66s" which does not match r"^.*[^0-9](?Pfailures[0-9]+) failed.*$" because there is nothing in front of the number.

This PR does a bit of refactoring to avoid this: Instead of creating summary_matches_pattern2 out of a RegEx match and using it in 2 places way below where the exact RegEx might be already forgotten I combined the counting of errors/failures with the summary-message generation and moved the RegEx right on top of the loop handling it and added comments with examples.

Also the anchors and named group is not required: r"^.*[^0-9](?Pfailures[0-9]+) failed.*$" -> r"([0-9]+) failed" which is much cleaner to read.

Finally my commit to also count tests terminated by a signal was lost so I readded it and reformatted the RegEx assignment to make it easier to read

And as a drive-by fix I added a docstring to get_count_for_pattern and made it return 0 instead of None for the case where the warning is shown.

[Flamefire]

Fixed yet another issue and tried to refactor it a bit to avoid more: Using finditer returns match objects which we can use to access RegEx groups by name.

Also in a PyTorch PR there was a syntax error in one of the tests but EB reported "Test command had non-zero exit code (1), but no failed tests found?!"

Similar if there were even a single test which failed then those syntax errors would have been undetected. I hence added my RegEx back which searches for failed test suites and if that doesn't match the ones we checked by the 2 custom regexes it will be reported as an error not taking the number of tests into account. I also include the failed test names in the error message for that case.

Edit: And finally fixed the minor spelling issue where it reported "0 failure" (missing "s")

[branfosj]

Test report by @branfosj

Overview of tested easyconfigs (in order)

• SUCCESS PyTorch-1.12.1-foss-2021a.eb

Build succeeded for 1 out of 1 (1 easyconfigs in total)
bear-pg0105u36b.bear.cluster - Linux RHEL 8.5, x86_64, Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz (icelake), Python 3.6.8
See https://gist.github.com/b52144157585d00a1b8d71ba060cb0a7 for a full test report.

[branfosj]

Going in, thanks @Flamefire!