PR summary

This PR fixes two bugs when flattening images for vector backends and alpha is not 1. Image flattening is when the Agg resampler is used to flatten a sequence of images into a single image so that only a single image needs to be drawn by the backend, and is on by default.

• When an image is added with a scalar alpha (imshow(..., alpha=...)), that scalar alpha is mistakenly applied twice: (1) when making the image to be resampled, and (2) when resampling the image into the common buffer. For example, an image that is intended to have an alpha of 0.6 ends up in the flattened image as if it had an alpha of 0.36 (0.6 squared). This is the root cause of Agg alpha image compositor differs from Matplotlib's software image compositor #8847.
• Filter images in premultiplied alpha mode. #29776 (and later fixed by Fixed inaccurate image placement and even more resampling bugs #31021) changed the Agg resampler to assume the input was in premultiplied alpha space, but the function composite_images had not been updated accordingly.

Fixes #8847

AI Disclosure

PR checklist

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