Re-implement cube repr_html

[pp-mo]

During #4778, found yet again that cube._repr_html_() was somewhat broken, and has been for some time.

This implementation is still problematic, mostly because it is based on re-parsing the output from str(cube).
What it should probably do is to work from a CubeSummary, not the printed text.
But this looks like a fair amount of work.

[pp-mo]

There are lots of tests,
yet none of them spotted the problems found above, even though they affect all cube-attribute + scalar-coordinate sections, which are very common in use.

There are tests in :

• tests/integration/experimental/test_CubeRepresentation.py
• tests/unit/experimental/representation/test_CubeListRepresentation.py
• tests/unit/experimental/representation/test_CubeRepresentation.py

For refactoring, it would be certainly be good to provide more integration tests, somehow.
The existing tests are definitely testing implementation (almost all private routine APIs), not behaviour.

A key problem is the difficulty of checking the complex html result
-- e.g. how would you catch the above-mentioned problem?
It could be easier if we used an object-structure intermediate, like maybe xml.etree, instead of just building strings.
Actually "lxml" might be a good choice, but I think it is a bit obscure ?
But it all means more work 😞

[github-actions]

In order to maintain a backlog of relevant issues, we automatically label them as stale after 500 days of inactivity.

If this issue is still important to you, then please comment on this issue and the stale label will be removed.

Otherwise this issue will be automatically closed in 28 days time.

[trexfeathers]

I still care about this

[github-actions]

In order to maintain a backlog of relevant issues, we automatically label them as stale after 500 days of inactivity.

If this issue is still important to you, then please comment on this issue and the stale label will be removed.

Otherwise this issue will be automatically closed in 28 days time.