fix: treat unmasked nans as bad when the input is a masked array

[matthiasschabel]

src/cmap/_colormap.py:409 picks the mask or the nan test, never both:

mask_bad = x.mask if np.ma.is_masked(x) else np.isnan(xa)

So a masked array's unmasked nans are never found. Such a value survives to xa.astype(int), becomes INT_MIN, and lut.take(..., mode="clip") clips it to index 0. The nan renders as the colormap's first ramp color, silently, and the result looks like ordinary data.

Colormap("viridis", bad="red") applied to np.ma.masked_array([0.0, 0.25, nan, 1.0], mask=[True, False, False, False]):

index	input	rendered	expected
0	0.0, masked	red	red
2	nan, not masked	(0.267, 0.005, 0.329), viridis at 0.0	red

The same four values as a plain ndarray render the nan red, so the two input types disagree about identical data. The docs say bad is the color "when values are NaN or masked" (docs/faq.md), so this restores the documented behavior.

Both conditions are needed to hit it: np.ma.is_masked returns False for an all-false mask, so those arrays already take the nan path and are fine.

Two details in the fix worth flagging:

• The nan test runs only when xa.dtype.kind == "f". A numeric object-dtype masked array works today precisely because it never reaches np.isnan, which raises on object dtype; the guard keeps that working.
• The masks combine with |, not |=, so the caller's own mask array is not written to. The test asserts that.

matplotlib 3.11 has the same line in Colormap._get_rgba_and_mask, comment included, so this is a divergence. It seems worth taking: the current output is not a decision about masked arrays, it is an integer-cast overflow clipped to index 0, and it contradicts cmap's own documentation of bad.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.57%. Comparing base (8040ef7) to head (2cc7653).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #148   +/-   ##
=======================================
  Coverage   95.56%   95.57%           
=======================================
  Files         168      168           
  Lines        2186     2190    +4     
=======================================
+ Hits         2089     2093    +4     
  Misses         97       97           

☔ 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.