Fix px.imshow ignoring zmin/zmax when the image contains NaN

[MohammadHijjawi97]

Link to issue

Closes #4299

Description of change

When only one of zmin/zmax is passed to px.imshow (or binary_string=True is used) with min/max contrast rescaling, the missing bound was computed with img.min()/img.max(), which return NaN if the image contains NaN values. The resulting NaN cmin/cmax made the user-provided bound have no effect. This uses np.nanmin/np.nanmax instead so NaN values are ignored.

Demo

import numpy as np, pandas as pd, plotly.express as px
fig = px.imshow(pd.DataFrame({"foo": [1, 2, 3, np.nan]}), zmax=2)
print(fig.layout.coloraxis.cmin, fig.layout.coloraxis.cmax)
# before: nan 2
# after:  1.0 2

Testing strategy

Added test_zmin_zmax_nan_data in tests/test_optional/test_px/test_imshow.py, which checks that passing only zmax or only zmin on an image containing NaN produces the expected cmin/cmax. It fails without the fix and passes with it; the rest of test_imshow.py is unchanged by this PR (on Windows, test_imshow_source_dtype_zmax[infer-int32] fails both before and after, in a branch this doesn't touch).

Guidelines

• I have reviewed the pull request guidelines and the Code of Conduct and confirm that this PR follows them.
• I have added an entry to the changelog if needed (not required for documentation PRs).