Instead of caching the text layout based on a bunch of properties, only
cache the computation of the text's metrics, which 1) should be the most
expensive part (everything else in _get_layout is relatively simple
iteration and arithmetic) and 2) depends on fewer implicit parameters.

In fact, the old cache key was insufficient in that it would conflate
usetex and non-usetex strings together, even though they have different
metrics; e.g. with the (extremely artificial) example

figtext(.1, .5, "foo\nbar", size=32)  # (0)
figtext(.1, .5, "foo\nbar", usetex=True, size=32, c="r", alpha=.5)  # (1)
figtext(.3, .5, "foo\nbar", usetex=True, size=32, c="r", alpha=.5)  # (2)

the linespacing of the first usetex string (1) would be "wrong": it is
bigger that the one of the second usetex string (2), because it instead
reuses the layout computed for the non-usetex string (0).

The motivation is also to in the future let the renderer have better
control on cache invalidation (with a yet-to-be-added renderer method),
e.g. multiple instances of the same renderer cache could share the same
layout info.

old:

new:

(compare the position of the left "foo")

(This is also orthogonal to #21958, which fixes another aspect of the text cache.)

PR Summary

PR Checklist

Tests and Styling

• Has pytest style unit tests (and pytest passes).
• Is Flake 8 compliant (install flake8-docstrings and run flake8 --docstring-convention=all).

Documentation

• New features are documented, with examples if plot related.
• New features have an entry in doc/users/next_whats_new/ (follow instructions in README.rst there).
• API changes documented in doc/api/next_api_changes/ (follow instructions in README.rst there).
• Documentation is sphinx and numpydoc compliant (the docs should build without error).