Confirm support for Python 3.13 and drop 3.8

[khaeru]

Python 3.13 will be released 2024-10-01 2024-10-07. This PR is to:

• Check/confirm that genno is compatible with the new version. This includes:
  • The existence of compatible releases of upstream dependencies. If those releases are not immediately available, the PR can aid to monitor and track upstream issues:
    • Python 3.13 error: cannot inherit frozen dataclass from a non-frozen one hgrecco/pint#1969 —blocking.
    • pyarrow 18.0.0 —see workaround below.
    • Python 3.13 support numba/llvmlite#1084 —for sparse only, currently addressed by omitting the install/tests of sparse in the CI matrix.
    • Python 3.13 numba/numba#9413 —for sparse only.
• Drop support for Python 3.8, which will reach end-of-life (no more updates, even security updates) "approximately [during, before, or after] October 2024" per PEP 569.
  • Remove any code marked as for Python 3.8 compatibility only. Depending on these changes, part or all of the functionality may remain usable with Python 3.8 if users have extreme need.
  • Make any general clean-ups or improvements possible with features available from Python 3.9.
    • Type hint using generics, e.g. list[str] instead of from typing import List/List[str].

PR checklist

• Checks all ✅
• Update documentation N/A
• Update doc/whatsnew.rst

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.85%. Comparing base (aeeda54) to head (54a23ff).
Report is 8 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #143      +/-   ##
==========================================
+ Coverage   99.80%   99.85%   +0.04%     
==========================================
  Files          51       51              
  Lines        4699     4691       -8     
==========================================
- Hits         4690     4684       -6     
+ Misses          9        7       -2     

Files with missing lines	Coverage Δ
genno/caching.py	100.00% <100.00%> (ø)
genno/compat/graphviz.py	100.00% <100.00%> (ø)
genno/compat/pint.py	100.00% <100.00%> (ø)
genno/compat/plotnine/plot.py	100.00% <100.00%> (ø)
genno/compat/pyam/__init__.py	100.00% <100.00%> (ø)
genno/compat/sdmx/operator.py	100.00% <100.00%> (ø)
genno/compat/sdmx/util.py	100.00% <100.00%> (ø)
genno/compat/xarray.py	100.00% <ø> (ø)
genno/config.py	100.00% <100.00%> (ø)
genno/core/attrseries.py	98.37% <100.00%> (ø)
... and 14 more

[khaeru]

@glatterf42 FYI, I am realising here that the hold-up pending pyarrow 18.0.0 is due to #135 / #128.

That change brought performance improvements, but introduced a fixed dependency on pandas[parquet] instead of merely pandas:
https://github.com/khaeru/genno/blame/18d3ec2c1e2929e4c1b7208ae7d17292680b6cba/pyproject.toml#L31

This in turn is blocking ixmp, message_ix, et al. and as you note there is no promised date for a release of pyarrow 18.0.0.

Because genno releases are easy to put out, I think I will temporarily make this:

  "pandas >= 1.0",
  "pandas[parquet] >= 1.0; python_version < '3.13'",

Thus on Python 3.13, parquet/pyarrow will not be installed, and caching should fall back to less-performant pickle from the standard library.

Once pyarrow 18.0.0 is out, I can make a point release of genno to revert the change.