FIX Clamp ImageResizingConverter output dimensions

[ThryLox]

Description

ImageResizingConverter can compute a zero width or height when a positive fractional scale factor is applied to a tiny image, causing Pillow to fail. It also accepted non-finite scale factors such as nan and inf.

This change rejects non-finite scale factors and clamps each computed output dimension to at least one pixel. Clamping allows tiny images to resize successfully rather than replacing Pillow's exception with another error, while preserving the existing int() truncation behavior whenever dimensions are already positive.

Fixes: #2162

Tests and Documentation

• Added validation coverage for nan, positive infinity, and negative infinity.
• Added resize coverage for 1x1 images and images with a one-pixel width or height.
• Ran the focused converter unit tests and file-scoped pre-commit checks.
• Documentation and JupyText updates are not applicable because the public API is unchanged.

[ThryLox]

@microsoft-github-policy-service agree

[rlundeen2]

Thx for taking a look! I was on the fence, but I think ultimately I'm going to close it. I know this is a judgement call, but I think this makes the code less readable; and ultimately it just changes the exceptions raised.

[romanlutz]

Thanks again for putting this together. I revisited the approach and would like to preserve this PR and your contribution. The non-finite scale-factor validation here is useful; for the zero-dimension case, the revised behavior should clamp each calculated dimension with max(1, int(...)) so tiny images resize successfully instead of replacing Pillow's exception with a custom one.

[romanlutz]

Updated this in a minor way to actually address the issue #2162

[akshan-main]

@romanlutz I thought it was good practice to open an issue before creating a PR, otherwise I would've created the PR directly myself. I feel like @ThryLox is taking credit for my finding. I'd have preferred to handle this.