BUG: Fix multiclass onehot inverse

[adriangb]

Resolves #188

One-hot encoded targets were not being processed correctly. The tests were checking for shapes and dtypes (which were correct), but the values were not.

This implements 2 things:

• 2 line fix for the bug in transformers.py
• Extensive test refactoring to test for estimator scores (to prevent this bug in the future).

[github-actions]

📝 Docs preview for commit f197d94 at: https://www.adriangb.com/scikeras/refs/pull/189/merge/

[codecov-io]

Codecov Report

Merging #189 (f197d94) into master (374b436) will increase coverage by 0.14%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #189      +/-   ##
==========================================
+ Coverage   99.55%   99.70%   +0.14%     
==========================================
  Files           6        6              
  Lines         671      669       -2     
==========================================
- Hits          668      667       -1     
+ Misses          3        2       -1     

Impacted Files	Coverage Δ
scikeras/wrappers.py	99.43% <ø> (ø)
scikeras/utils/transformers.py	100.00% <100.00%> (+0.93%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 374b436...f197d94. Read the comment docs.

[adriangb]

#188 (comment) confirms this fixes the original bug report. Will move forward with merge.