perf: Experiments in trying to speed up CI

[itamarst]

• sccache instead of normal all-or-nothing GHA cache
• Give test python builds their own profile, since it seems like they were never getting cached at all. My guess is that custom RUSTFLAGS meant the code coverage build was trashing any relevant cache.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.17%. Comparing base (4fc5929) to head (f97bd1d).
Report is 27 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #21010      +/-   ##
==========================================
- Coverage   79.20%   79.17%   -0.04%     
==========================================
  Files        1583     1583              
  Lines      225105   225255     +150     
  Branches     2581     2586       +5     
==========================================
+ Hits       178294   178337      +43     
- Misses      46221    46328     +107     
  Partials      590      590              

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

[itamarst]

Notes for when I return to this next week:

• Unclear if sccache does anything.
• Why do test-python jobs appear to recompile everything? Was in the middle of checking if maturin develop -m py-polars/Cargo.toml made a difference, that's what the coverage job does (that one doesn't recompile everything?!).
• adding pytest -n auto to io tests was useful. Next step is forcing rebuild, ideally with no code changes.
• why is uv on windows slow?
• check if --verbose on uv pip install is slowing things down.
• check if tests are sufficiently faster with higher optimization level to merit switching.