Add string comparison support so that x.sort() works

[peytondmurray]

This PR builds on numpy/numpy#23173 to provide a PyArray_CompareFunc so that x.sort() works.

I also updated the README with a bit of information about needing the nightly numpy wheel for support, added a note about --no-build-isolation, and removed a bit about meson-python not supporting editable builds (it does now!).

This PR remains in draft until numpy/numpy#23173 is merged.

[ngoldbaum]

Can we get a test too?

[ngoldbaum]

It occurs to me that it may not be correct to rely on strcmp to reliably sort utf-8 encoded data in a sensible way. It may be a good idea to look up edge cases related to sorting unicode strings to see how this approach handles them.

[mattip]

Unicode string comparison is tricky. There are lots of edge cases.

[peytondmurray]

Yep, you're both right - strcmp will work for the ASCII dataset but a more robust approach seems important here. I'll spend some time looking into this.

[peytondmurray]

After some discussion I've made the choice to sort strings by their UCS4 code points. This is the approach that python takes by default, and if you want the more advanced sorting methods required for unicode comparison, you can use the unicodedata package.

I still think it would be cool to have vectorized functions for normalization and sorting, but we can introduce those later, since you can't provide the options you would need to do that for arr.sort() here anyway.

[ngoldbaum]

Besides the inline comment, can you also increment the API version in the other dtypes? You'll need to do that anyway before this is merged to get the tests to pass.

[peytondmurray]

Thanks for the feedback - I've updated the API versions everywhere.