RFC: refactor plot callbacks registration

[neutrinoceros]

PR Summary

Fix #3945
Right now this is a proof of concept, there's still work to do, but I think it's promising so I'm opening to see how many tests are currently broken.
It should backwards compatible, and retain compatibility with externally defined plot callbacks (such as HaloCatalogCallback from yt_astro_analysis).

If I can manage to get it 100% complete while keeping a reasonable diff size, I'll propose this for yt 4.1

Note: I ended up cleaning up a lot of signatures in PlotCallback subclasses __init__ methods (making most arguments keyword only). This change is technically not backward compatible. It is motivated by future compatibility concerns: some arguments are already deprecated, but sometimes occupy early positions, so there is no way to actually remove them in the future without breaking backward compatibility. Making them keyword only now is the least painful route I can see.
This wasn't the original goal of this PR, however it turned out that to be a requirement to keep this as tidy as possible, I'm happy to discuss it in more detail upon request.

In short, here's what works on the main branch but not here:

import yt

ds = yt.load_sample("IsolatedGalaxy")
p = yt.SlicePlot(ds, "z", ("gas", "density"))
p.annotate_contour(("gas", "density"), 5, 4, (1e-28, 1e-25))
p.save("/tmp/test_contour.png")

on this branch it will raise a TypeError (this is 100% by design)

Traceback (most recent call last):
  File "/Users/robcleme/dev/yt-project/yt/t_callback_validation.py", line 5, in <module>
    p.annotate_contour(("gas", "density"),5,4, (1e-28, 1e-25))
  File "/Users/robcleme/dev/yt-project/yt/yt/visualization/plot_modifications.py", line 103, in closure
    self._callbacks.append(cls(*args, **kwargs))
TypeError: ContourCallback.__init__() takes from 2 to 3 positional arguments but 5 were given

The fix is simply to use keyword arguments instead of positional:

p.annotate_contour(("gas", "density"), levels=5, factor=4, clim=(1e-28, 1e-25))

(in this particular example, the second argument, levels, main still be passed positionally)
Hopefully this is not too concerning of a change, since most arguments were desinged to be used as keyword rather than positional, and that's how they've always been demonstrated in docs. The main reason why they were not already implemented as keyword only is that this wasn't a thing in Python 2, but it's been around since Python 3.0

TODO:

• proof of concept for PlotCallback subclasses
• generalize to FRB filters
• validate CI
• move geometry checking responsability from __call__ to __init__
• reimplement mechanism to detect unsupported callbacks according to plot type
• check that annotate_halos from yt_astro_analysis still works as expected (checked with https://yt-astro-analysis.readthedocs.io/en/latest/halo_plot_callbacks.html?highlight=annotate_halo#annotate_halos)

PR Checklist

• [N/A] New features are documented, with docstrings and narrative docs
• Adds a test for any bugs fixed. Adds tests for new features.

[neutrinoceros]

note that this registry becomes less relevant with this patch, since we don't rely on it to setup annotation methods.
However it's still used in some places (most notably in a docs helper script), so I'm keeping it.

[matthewturk]

seems OK, but also lots more typing fixes than I'd have expected!

[neutrinoceros]

Not really fixes actually, mypy wasn't complaining about their absence. I just figured I might as well encode what I learned while I was at it :)

[brittonsmith]

I'm happy with this as is. Very nice refactor!

[neutrinoceros]

Thanks ! anyone wants to merge ?