FIX: gridliner auto-updates by default (and always in future)

[rcomer]

Rationale

As suggested at #1792 (comment), updating gridlines and their labels when the figure is re-drawn (or tight_bbox re-calculated) is now the default. Also deprecate the option to not update.

Closes #1792

Implications

There is one change in an image test result, where the zero labels on the equator have disappeared from the top right plot. I'm not 100% clear why this is. There is some logic that makes labels invisible based on how much space is available so my best guess is that the calculation after tight_layout puts us just the other side of the threshold for "not enough space", though any changes in positioning are imperceptible.

I guess if anyone tells us they have a use-case where auto_update=False matters, we can remove the deprecation.

[rcomer]

Is there a way to view the failed test images from GHA? I do not have access to a Windows machine. I note that the RMS on the failures is similar to what this test's tolerance was previously set to, so it's probably fine but better to check!

[greglucas]

Is there a way to view the failed test images from GHA? I do not have access to a Windows machine. I note that the RMS on the failures is similar to what this test's tolerance was previously set to, so it's probably fine but better to check!

If you click on the "Summary" page for Actions on all the runners you can see an "Artifacts" download area then. Here is a direct link:
https://github.com/SciTools/cartopy/actions/runs/8008180591?pr=2330#artifacts

[rcomer]

Thanks @greglucas - here is the failed diff from Windows py311:

Just a small shift in the location of two zeros. So I think we are fine to just set the tolerance as before 👍

[rcomer]

If I pass the auto_update parameter under 0.23, I don't see a deprecation warning. What did I do wrong here?

[lgolston]

If I pass the auto_update parameter under 0.23, I don't see a deprecation warning. What did I do wrong here?

With warnings.simplefilter('always') the warning does show up. Apparently FutureWarning needs to be used instead of DeprecationWarning to show by default.

[lgolston]

After reading https://peps.python.org/pep-0565/, I think that is not actually the problem. Testing now, if stacklevel=3 is added to the warning then it does show