PR summary

Due to floating-point precision, a figure may self-report a height that is slightly below an integer (by one floating-point tick). When this happens, there are (at least) two bugs:

• The reported mouse position in GUI events is inconsistent between pixel coordinates and data coordinates (see [Bug]: QuadMesh.get_cursor_data() returns incorrect value #27570).
• The rendered Agg/Cairo output is too short by a pixel (see also Reliably set the output image size #15363):

>>> import matplotlib.pyplot as plt
>>> fig = plt.figure(figsize=(1, 2.03), dpi=100)
>>> print(*fig.bbox.size)
100.0 202.99999999999997
>>> fig.canvas.draw()
>>> print(fig.canvas.buffer_rgba().shape[1::-1])
(100, 202)

This PR:

• Fixes FigureCanvasBase.get_width_height() so that it accounts for floating-point precision before truncating to integers so that the canvas will be the intended size.
• Switch backends to expressly request the canvas size in integral pixels (where appropriate) rather than internally calculate it from the figure size.

Fixes #27570
Fixes #15363

AI Disclosure

No AI was used

PR checklist

• "closes #0000" is in the body of the PR description to link the related issue
• new and changed code is tested
• [N/A] Plotting related features are demonstrated in an example
• [N/A] New Features and API Changes are noted with a directive and release note
• [N/A] Documentation complies with general and docstring guidelines