Fix multiclass macro F-score (mean of per-class F, not F of the means)

[gaoflow]

Multiclass f_score / f1_score returned the F-beta of the macro-averaged precision and recall, but the macro F-score is the mean of the per-class F-beta scores. Those agree only when every class shares the same precision and recall, so multiclass f1_score was wrong for most inputs.

Repro (y_true=[0,0,1,1,2,2], y_pred=[0,1,1,1,2,2]): f1_score() returned 0.860215; the macro F1 is 0.822222 (per-class F1 = [2/3, 4/5, 1], matching sklearn f1_score(average="macro")). Macro precision and recall were already correct, which isolates the error to the F combination.

The fix mirrors the existing precision/recall macro averaging: average the per-class one-vs-all F-beta. A class the model never gets right has precision = recall = 0 (a 0/0 F); it now counts as 0 (matching sklearn's zero_division=0) instead of turning the whole average into NaN. Binary matrices are unchanged.

Two regression tests added; full cargo test --workspace, cargo fmt and cargo clippy pass.

One caveat: for beta != 1 the aggregation is now correct, but linfa's per-class precision and recall are oriented opposite to sklearn's, so f_score(beta) lines up with sklearn's fbeta_score(beta=1/beta). That orientation is pre-existing (it affects the binary path too) and left out of scope here; f1 is symmetric so it is unaffected.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.92%. Comparing base (7fe5c86) to head (26d20db).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #448      +/-   ##
==========================================
+ Coverage   77.53%   77.92%   +0.39%     
==========================================
  Files         106      104       -2     
  Lines        7585     7534      -51     
==========================================
- Hits         5881     5871      -10     
+ Misses       1704     1663      -41     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.