ENH: Add screen-proportional mutation_aspect to FancyBboxPatch - #32205

PR summary

What problem existed

FancyBboxPatch requires a numeric mutation_aspect to compensate for a non-unit axes aspect ratio. In a plot with aspect="auto" (e.g. a typical bar chart), the rounding of BoxStyle("Round", ...) is drawn as an ellipse on screen, not a circle. The only way to get isotropic rounding was the private workaround mutation_aspect=1/ax._get_aspect_ratio(), which is also computed too early (before layout) and therefore stale by draw time.

How it was reproduced

import matplotlib.pyplot as plt
from matplotlib.patches import FancyBboxPatch, BoxStyle

def add_bar(ax, x, y, w, h, mutation_aspect=1):
    patch = FancyBboxPatch(
        (x, y), w, h,
        boxstyle=BoxStyle("Round", pad=0, rounding_size=w / 2),
        linewidth=0, facecolor="C0",
        mutation_aspect=mutation_aspect,
    )
    ax.add_patch(patch)

fig, ax = plt.subplots(figsize=(10, 5))
x, y, w, h = 0.3, 0, 0.18, 1
add_bar(ax, x, y, w, h)
ax.autoscale_view()
fig.canvas.draw()

What caused it

FancyBboxPatch.get_path() applies mutation_aspect in data coordinates only. When the axes display scaling differs between x and y (aspect != 1), a data-space circle is drawn as an on-screen ellipse. There was no public way to opt in to resolving the aspect ratio from the axes.

What changed

FancyBboxPatch.mutation_aspect now additionally accepts the string "screen-proportional". At draw time the patch resolves it from the owning axes using public API (get_window_extent, axis transforms), so the box is scaled like the data on screen and the rounding renders circular regardless of the axes aspect ratio. set_mutation_aspect validates the new value.

Why this approach

The issue proposed an opt-in mode; discussion suggested a name that is more specific than a generic "auto". The resolution is deferred to draw time so the layout (and thus the display scaling) is final, fixing the stale-value problem of the private-API workaround. All existing numeric behavior is unchanged.

How it was tested

• python -m pytest lib/matplotlib/tests/test_patches.py -k "mutation_aspect" -v → 1 passed
• python -m pytest lib/matplotlib/tests/test_patches.py → 73 passed, 11 skipped (pre-existing skips)
• Added tests: display-space circular corner check with a non-unit-aspect axes, fallback to 1 without an Axes, and ValueError on invalid values.

Risks / limitations

• Without an attached Axes, "screen-proportional" falls back to an aspect of 1 (documented in the new tests).
• The resolved aspect is a snapshot at draw time; it updates when the figure is redrawn after a resize.

Related issue

Closes #31175