After bouncing some ideas off the wgpu chat, the conclusion is to rewrite the shader to use multiple samplers and do the individual scaling steps. I don't want to prematurely optimize, so it's going to do both samplers scaling steps regardless of the PAR value.

The next checklist item would be an API to allow changing the PAR at runtime. I don't know if this will be useful, but it's always possible that someone comes along with a use case for it. It would also be an improvement for the PAR example; a key to toggle PAR correction on/off will be more illustrative than just some static screenshots in the README.

Sorry to be a nag but what is the state of this PR? Anything we can do to help?

There aren't any updates for this PR other than what is already here. Specifically, the extra context provided in #262 kind of obsoletes the implementation in this PR. In short, this does allow for non-square pixel aspect ratios, but it comes at the cost of "pixel shimmering" in animations, especially when things move horizontally. Texture filtering like the two-pass implementation in #262 can address that problem.

There is still an unresolved issue there; how to integrate the second pass into the library as a first-class feature (that PR currently requires it to be added to a project as a custom renderer). If you are interested in helping drive the effort, that would be the place to start.