Raise ValueError for zero or negative sizes in ImageOps contain/cover/pad/fit

[hdimer]

ImageOps.contain, cover, pad and fit handle invalid output sizes inconsistently. Because they compute size[0] / size[1] before Image.resize validates the size, a zero height raises ZeroDivisionError; cover skips validation entirely for a zero or negative width (returning a wrong-sized image); and contain returns a wrong-sized image for a negative height.

Before, on a 100x50 image:

size	contain	cover	pad	fit
(0, 10)	ValueError	(20, 10)	ValueError	ValueError
(10, 0)	ZeroDivisionError	ZeroDivisionError	ZeroDivisionError	ZeroDivisionError
(-1, 10)	ValueError	(20, 10)	ValueError	ValueError
(10, -1)	(10, 5)	ValueError	ValueError	ValueError

This adds an up-front size check to contain, cover and fit (pad delegates to contain) that raises the core's own ValueError: height and width must be > 0 for any dimension <= 0, so all four behave consistently.

Testing

Added a parametrized test over the four functions and the zero/negative sizes above; it asserts a ValueError rather than a crash or a silently wrong result. The full Tests/test_imageops.py suite passes (62 tests) and ruff is clean. A release note is included.

---

Used AI assistance on this; I reviewed and tested the change myself.

[radarhere]

This topic has had some recent activity in other PRs.

#9672 suggested resolving the contain() ValueError by returning images of at least 1px.
In response, I created #9673, suggesting successfully returning an image where one dimension is zero.

I may investigate expanding #9673 to address these other operations.