Fix pandas 3 data type and copy/write bugs

[maltekuehl]

Reference Issue or PRs

See: https://scverse.zulipchat.com/#narrow/channel/316218-repo-management/topic/Upcoming.20Pandas.203.2E0.2E0.20release/with/569443597

What does your PR implement? Be specific.

Fixes pandas 3 (now released) incompatibility due to bool to float conversion & update on copy of DataFrame errors as well as explicit keyword arguments on mean/sum.

• Version bumped to 0.5.4, as there should be no incompatibilies.
• Changed all instances of .sum(0)/.mean(0)/.sum()/.mean() to `.sum(axis=0)/.mean(axis=0)
• Initialized convergence tracking columns with nullable boolean pd.array instead of float64 np.array with np.nan, so that bool assignments are possible without conversion / error
• Changed LFC shrinkage update, since in pandas 3 it would update a copy (views are now copies) instead of the original df. Since there is the theoretic possibility of SE/LFC being NaN or inifinite after shrinkage, implemented a isfinite check to match previous behavior of only updating on non-NaN values. However, since I personally find this to be unexpected behavior (I would not expect lfc_shrink to silently perform partial shrinkage), I have now added a warning informing users when partial shrinkage occurs (should be almost never, but I cannot exclude that this cannot happen).
• I have build the docs & compared all LFC array prints in the notebooks to the previous version and they match 1:1 and all tests are passing.

[grst]

LGTM, I have two unrelated comments -- not necessary to fix in this PR.

[maltekuehl]

LGTM, I have two unrelated comments -- not necessary to fix in this PR.

Thanks! Agree with the proposed changes, but outside the scope of what I can allocate time to this week. Since it's currently broken with pandas 3, probably best to raise an issue for these points for a follow up PR.