Gridliner: don't destroy and recreate artists

[rcomer]

Rationale

@greglucas has asked for this at least twice!

When re-drawing the GridLiner, update its existing LineCollection and Text objects rather than creating new ones. This was trivial for the LineCollections but a bit more complicated for the labels since we might have different numbers of them e.g. after zooming. Clearly Matplotlib's Axis has already solved that problem so I got my inspiration from there. We now keep two lists of labels: _all_labels is the list of labels that have been drawn at some point. _labels is the list that were used at the most recent draw (and still gets cleared out and repopulated each time).

Goes on top of #2249. Only the third commit is new.

Implications

[rcomer]

Putting this one in draft as it will need rebasing once #2249 is merged.

[greglucas]

I tested this out locally zooming and panning around and it seems quite responsive to me, nice work!

One minor pre-commit failure to fix before merging.

[rcomer]

Apologies, I don't know how I managed to push that failure as I have pre-commit installed locally. I should have checked the CI myself though.

[rcomer]

I broke the minimum versions tests. In fact I broke them at #2249: Figure.draw_without_rendering and the layout keyword for figure, etc. were introduced at mpl v3.5. Should I fix them all here or make a separate PR?

[rcomer]

I note that mpl v3.5 is two years old next month, so the easiest fix would be to move the pin.

The problem calls are only in the tests.

[greglucas]

I broke the minimum versions tests. In fact I broke them at #2249: Figure.draw_without_rendering and the layout keyword for figure, etc. were introduced at mpl v3.5. Should I fix them all here or make a separate PR?

:( I should have checked that during the review but I was only looking at the ones without min-version specified. That is the problem when you have external issues causing red x's throughout the test suite.

I'd say it should probably be a separate commit, but I'm fine having it in this PR if that is easier than a new PR.

I note that mpl v3.5 is two years old next month, so the easiest fix would be to move the pin.
The problem calls are only in the tests.

I'd say that bumping the pin is fine then. We are not going to release within the next month. Note that you should also probably remove any other MPL3_4 checks there are (if any) and remove that code too, and if so then a new PR might be worthwhile.

[dopplershift]

👍 to what @greglucas said above.

[rcomer]

I decided to put it in a separate PR #2258

[rcomer]

I think this one is now ready to go. The only change since @greglucas' approval is the flake8 fix and I have also checked that all the test runs have exactly ten failures.

[greglucas]

Still looks good to me. I verified the test runs as well.