Please see the comment at #29368 (comment). I think we should go further and discourage its use to prevent a maintenance burden of a method that doesn't have much advantage.

Discouraging/deprecating is more than I want to invest into this PR. I just realized, that we have not documented what non-uniform actually means and added a quick description. IMHO that's a plus even if we decide to discrouage/deprecate.

Discouraging/deprecating needs a dedicated discussion. There are some things to consider:

• There's some existing usage https://github.com/search?q=%2FNonUniformImage%5C%28%5Ba-z%5D%2F&type=code
• NonUniformImage has recently seen a bugfix BUG: Fix NonUniformImage with nonlinear scale #27964
  So at least there's some interest in it (not judging whether that's justified or people should use something else). This also means one has to clearly describe the alternatives. That's not a priority of mine for now. For simplicity, I'd leave NonUniformImage as is. But you are welcome to push the topic.

@dstansby I've accepted your proposal. Please squash-merge.