test_Data workflow with skips to monitor Dask migration

[sadielbartholomew]

I am about to put up some more interesting PRs to make headway with 'Daskifying' the data-related methods in turn for #182, but before I start doing that in earnest I want to ensure we can track the migration progress in terms of the tests that pass, notably for (self-)review to:

• indicate progress for a given PR by highlighting extra tests that pass due to a given change; and
• ensure tests that do pass due to progress previously made aren't broken by any proposed changes in a new PR.

So in this PR I have:

• added skips across test_Data to mark tests not yet made to pass, that can all be switched on or off by a new flag TEST_DASKIFIED_ONLY;
• added a new workflow specifically specifically for the LAMA-to-dask migration work to trigger Github Actions to run jobs that run the test_Data module and not the whole test suite, because the first goal is to get the data modules working, where metadata tests will fail on data issues until that is complete.

[sadielbartholomew]

The new workflow clearly didn't trigger, I try to get it to start on Monday and go from there.

[sadielbartholomew]

Using an open-close to see if that will result in a trigger of the new workflow...

[sadielbartholomew]

Triggering worked 🎉 With the new test skips, I want to get the jobs to pass before I can consider the new workflow to be ready, so let's see how they fare...

[sadielbartholomew]

Most of the Ubuntu jobs have passed, largely due to the intentional skips on the test_Data they ultimately run:

...

----------------------------------------------------------------------
Ran 73 tests in 0.221s

OK (skipped=66)

...

with the Mac OS jobs generally taking longer and being cancelled by the one failure on the Ubuntu with Python 3.6 job. I have just checked and Dask no longer supports Python 3.6 (https://github.com/dask/dask/blob/main/setup.py#L71), hence the lone test failure on test_Data___setitem__ for that version only due to a NotImplementedError: Item assignment with <class 'tuple'> not supported. I will remove Python 3.6 from the job matrix in the next commit, and hopefully both the Ubuntu and Mac OS jobs will then all pass.

[sadielbartholomew]

Re-triggering via open-close...

[sadielbartholomew]

Sorry @davidhassell I initially added you to review and then realised there is nothing much to review here so removed you as a reviewer, but please do read the opening comment so you know about the skipping of some test_Data test methods which will be incrementally un-skipped as more progress is made, and the new workflow to run test_Data only for PRs to the lama-to-dask development branch for #182.