fix nan mismatch handling in tolerance

[jdries]

With this fix, pixels with diff in NaN will be added to tolerance.
Avoids false positives in apex testing.
Example of improved message when tolerance is 0:

   raise AssertionError("\n".join(issues))
E           AssertionError: Issues for variable 'var':
E           var: Nodata (NaN) mismatch: 1/582267 pixels (0.0001717% > 0%): 1 NaN only in actual, 0 NaN only in expected
E           Left and right DataArray objects are not close
E           Differing values:
E           L
E               array([[[0.068862, 0.167273, ..., 0.235385, 0.016265],
E                       [0.097264, 0.107955, ..., 0.157202, 0.057394],
E                       ...,
E                       [0.075301, 0.105344, ..., 0.055046, 0.142601],
E                       [0.100671, 0.090909, ..., 0.128065, 0.152412]],
E               
E                      [[     nan,      nan, ...,      nan,      nan],
E                       [     nan,      nan, ...,      nan,      nan],
E                       ...,
E                       [     nan,      nan, ...,      nan,      nan],
E                       [     nan,      nan, ...,      nan,      nan]],
E               
E                      ...,
E               
E                      [[0.10089 , 0.084615, ..., 0.13969 , 0.092689],
E                       [0.079595, 0.086667, ..., 0.204738, 0.026182],
E                       ...,
E                       [     nan,      nan, ..., 0.181572, 0.067144],
E                       [     nan,      nan, ..., 0.115353, 0.170088]],
E               
E                      [[     nan,      nan, ...,      nan,      nan],
E                       [     nan,      nan, ...,      nan,      nan],
E                       ...,
E                       [0.061889, 0.005425, ...,      nan,      nan],
E                       [0.03645 , 0.065217, ...,      nan,      nan]]],
E                     shape=(7, 233, 357), dtype=float32)
E           R
E               array([[[0.068862, 0.167273, ..., 0.235385, 0.016265],
E                       [0.097264, 0.107955, ..., 0.157202, 0.057394],
E                       ...,
E                       [0.075301, 0.105344, ..., 0.055046, 0.142601],
E                       [0.100671, 0.090909, ..., 0.128065, 0.152412]],
E               
E                      [[     nan,      nan, ...,      nan,      nan],
E                       [     nan,      nan, ...,      nan,      nan],
E                       ...,
E                       [     nan,      nan, ...,      nan,      nan],
E                       [     nan,      nan, ...,      nan,      nan]],
E               
E                      ...,
E               
E                      [[0.10089 , 0.084615, ..., 0.13969 , 0.092689],
E                       [0.079595, 0.086667, ..., 0.204738, 0.026182],
E                       ...,
E                       [     nan,      nan, ..., 0.181572, 0.067144],
E                       [     nan,      nan, ..., 0.115353, 0.170088]],
E               
E                      [[     nan,      nan, ...,      nan,      nan],
E                       [     nan,      nan, ...,      nan,      nan],
E                       ...,
E                       [0.061889, 0.005425, ...,      nan,      nan],
E                       [0.03645 , 0.065217, ...,      nan,      nan]]],
E                     shape=(7, 233, 357), dtype=float32)
E           var: t 2025-05-19T00:00:00.000000000: nodata (NaN) mismatch: 1/83181 pixels (0.0%)