Handle --per-contig grouping with no gene tags

[akmorrow13]

Replaces https://github.com/CGATOxford/UMI-tools/pull/570, adds unit test

[akmorrow13]

Can one of the admins verify this patch?

[akmorrow13]

@TomSmithCGAT can you trigger these tests?

[TomSmithCGAT]

@akmorrow13 - Thanks for the prompt

Please ignore the error. That's because we are using nose, which is incompatible with python 3.10 (https://github.com/nose-devs/nose/issues/1099 & https://github.com/CGATOxford/UMI-tools/pull/546) and we appear to have added py 3.10 to our testing when removing py 3.6 (https://github.com/CGATOxford/UMI-tools/commit/8a7c9c43d7c18fa0d7a82af62782023f7ad164e7).

I'll fix the tests on master and then re-run here.

[akmorrow13]

@TomSmithCGAT thank you for your help. There was a bug in my test, it should now be fixed.

[TomSmithCGAT]

OK, this is very strange. With the above further commits, I can get the new test to pass with py3.7 & 3.8 but not 3.9 😖

I've can replicate this behaviour locally by switching python versions, so it does seem to be something inherently different between the python versions. I understand dictionaries output order changed recently, but that should all be python>=3.6.

Leave this with me and I'll try and work out what's going on.

[TomSmithCGAT]

🎉 Test passing

@IanSudbery - Whatever the difference is between py3.8 & 3.9, it manifests as differences in the output order. No reason for our tests to fail for different order. Previously, we had a 'Sort' flag for tests where the stdout should be sorted prior to comparison with reference. I've extended this so that it sorts all reference and output files prior to comparison. We really should switch to a more standard test framework when one of has some spare time!

Are you happy for me to merge this?

[akmorrow13]

Thank you so much for all the work on this @TomSmithCGAT!

[IanSudbery]

Looks good, go ahead.