ENH: add support for symlog colorbars with arbitrary bases (requires MPL>=3.5)

[neutrinoceros]

PR Summary

replay #4447 + closes #4448

While the set of minor ticks on a log scale isn't well defined in the general case (say, natural log), I find that our existing approach for log10 (get_symlog_(minor|major)ticks()) can be generalized to work with any integer base. This is essentially what I'm doing here, using pure MPL API. For non-integer bases, I delegate to matplotlib (which is equivalent of giving up minor ticks).

As noted in #4448, I found it was very hard (likely impossible) to fix the issue with matplotlib older than 3.5, so for these versions, I'm keeping bug-for-bug compatibility.

TODO:

• validate image tests (or be satisfied with differences)
• add a test
• resolve related discussions upstream
  • [Bug]: (edge case) no ticks are drawn in colorbars with SymLogNorm  matplotlib/matplotlib#25945
  • [ENH]: could there be a long-term stable API to access the log base from a SymLogNorm object ? matplotlib/matplotlib#25946
• adapt to any upstream change resulting from those discussions
• clean branch history
• merge companion PR TST: update image baseline in support to yt PR 4449 yt_pytest_mpl_baseline#3

I also opened matplotlib/matplotlib#25994 but the further I dig in, the less likely it looks that this will have a swift, short term solution upstream, so I'm not making its resolution a requirement for this PR.

[neutrinoceros]

Still having 4 errors in pytest-mpl tests, 3 of which are identical and all of which result from leaning more on matplotlib to determine major ticks in symlog, and only affect edge cases. I believe these differences are acceptable (desirable, even).

Next step will be to open a discussion with matplotlib devs to see which part of this work (if any) they'd be interested in upstreaming.

[matthewturk]

This looks good to me, and great work proposing it for upstreaming.

[neutrinoceros]

pre-commit.ci autofix

[neutrinoceros]

Turns out merely keeping bug-for-bug compatibility for MPL<3.5 adds a lot of complexity to ImagePlotMPL._set_axes, which IMO hurts maintainability and probably makes reviewing this PR much harder. It would make my life much easier to just require 3.5 everywhere.

[neutrinoceros]

Alright this is now stable enough for review. All remaining image failures are expected and intended, the companion PR for the baseline submodule may be found at yt-project/yt_pytest_mpl_baseline#3

[neutrinoceros]

@matthewturk the companion PR should have been merged first ! I'll open a follow up PR to fix tests on main