ImageOps.contain and pad fail when a narrow image rounds to zero pixels

Resizing a narrow image with ImageOps.contain() can round the shorter dimension to zero, even when both requested dimensions are positive. ImageOps.pad() hits the same error because it calls contain().

from PIL import Image, ImageOps

image = Image.new("RGB", (100, 1), "red")
ImageOps.contain(image, (10, 10))
# ValueError: height and width must be > 0

ImageOps.pad(image, (10, 10))
# ValueError: height and width must be > 0

I would expect contain() to return a 10×1 image, and pad() to return that image within a 10×10 canvas. The transposed input, 1×100, fails too. A 20×1 input also fails at the half-pixel rounding boundary.

In contain(), the calculated dimension is passed directly from round() to resize(). Keeping that calculated dimension at least one pixel avoids the error.

Reproduced on Windows with Python 3.13.15 and Pillow 12.3.0, and with ImageOps.py from main at 0a61c530aed1a2792c6a47674ac2b70753bfccbb loaded against the same installed Pillow core. No external image or network access is needed for the reproduction.

Prepared with OpenAI Codex assistance.

Start in ImageOps.py at ImageOps.contain(), where the rounded calculated dimension is passed to resize(); inspect how ImageOps.pad() delegates to contain(). Reproduce the narrow and transposed cases from the issue, then verify that contain() preserves a one-pixel dimension and pad() places the result in the requested canvas without the zero-size error.