Fix NaN handling in electrostatic_sphere_eb MR analysis (#6807)

## Summary

In IEEE floats, `np.nan != np.nan` is _always_ `True`, so the old code
was buggy:
NaN is never equal to anything, _including itself_.

## Original Description - Spotted by @zippylab

The find_first_non_zero_from_bottom_left/upper_right helper functions
used ``matrix[i][j] != np.nan`` to skip NaN values. Per IEEE 754, NaN !=
NaN is always True, so this check never filtered out NaN cells.

On GPU platforms where openpmd outputs NaN for cells outside the refined
patch (rather than zero), the analysis selected NaN regions as valid
data, producing ``nan`` errors for level 1 and failing the assertion.

Fix: replace ``!= np.nan`` with ``not np.isnan()``.

With the fix, level 1 errors on A100 GPU are 0.029% (phi) and 0.083%
(Er), well within the 0.4% tolerance.

## X-ref

Isolated from #6801

Co-authored-by: Tim Williams <tjwilliams@anl.gov>

Author: ax3l

Examples/Tests/electrostatic_sphere_eb/analysis_rz_mr.py

@@ -26,15 +26,15 @@
 def find_first_non_zero_from_bottom_left(matrix):
     for i in range(matrix.shape[0]):
         for j in range(matrix.shape[1]):
-            if (matrix[i][j] != 0) and (matrix[i][j] != np.nan):
+            if (matrix[i][j] != 0) and (not np.isnan(matrix[i][j])):
                 return (i, j)
     return i, j
 
 
 def find_first_non_zero_from_upper_right(matrix):
     for i in range(matrix.shape[0] - 1, -1, -1):
         for j in range(matrix.shape[1] - 1, -1, -1):
-            if (matrix[i][j] != 0) and (matrix[i][j] != np.nan):
+            if (matrix[i][j] != 0) and (not np.isnan(matrix[i][j])):
                 return (i, j)
     return i, j