compat: Numba 0.61 and Python 3.13

[hoxbro]

Windows does not solve because of libxml2 pin and Python 3.13 because of zlib pin.

numba/llvmlite#1105

test-gpu:

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.36%. Comparing base (f23de59) to head (f447818).
Report is 6 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1373      +/-   ##
==========================================
- Coverage   88.53%   86.36%   -2.17%     
==========================================
  Files          93       93              
  Lines       18744    18744              
==========================================
- Hits        16595    16189     -406     
- Misses       2149     2555     +406     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[amaloney]

I must be doing something wrong as I am unable to solve an environment using pixi. I'm running pixi shell in order to create the default environment, but instead of creating a default one, it attempts to create all environments in the environments table and will not solve anything unless I add the platform to the features.test-gpu table.

[hoxbro]

I'm running pixi shell in order to create the default environment

This PR is still in draft; I just wanted to see if PyPi would solve it, which it wouldn't. What you are seeing is the solving of environments with PyPi dependencies, where it needs to download the PyPI package to solve. I have removed this and will wait on the conda-forge release.

[hoxbro]

I don't think we can wait anymore for the release of numba and llvmlite on conda-forge for releasing datashader 0.17.0. Currently, it needs conda-forge/llvmlite-feedstock#96 to be merged, followed by conda-forge/numba-feedstock#150.

Most changes to pixi.toml can be reverted when numba is released on conda-forge. See NOTE: comments.

@amaloney: I see two options for this PR: merge the PR as is or do a new PR with the updates to README.md and pyproject.toml classifiers. I'm leaning toward the latter.

[hoxbro]

superseded by #1403