RandGridPatch crashes for documented 0 or None patch_size entries

Describe the bug
RandGridPatch documents that patch_size entries of 0 or None select the whole dimension, matching GridPatch. However, when max_offset is not supplied, RandGridPatch.__call__ computes s % p for each spatial dimension. This crashes for p == 0 and p is None before patch generation starts.

The dictionary wrapper RandGridPatchd is affected through the same array transform.

To Reproduce

import torch
from monai.transforms import GridPatch, RandGridPatch

img = torch.arange(1 * 8 * 8 * 8, dtype=torch.float32).reshape(1, 8, 8, 8)

for patch_size in [(0, 4, 4), (None, 4, 4)]:
    print("GridPatch", patch_size, GridPatch(patch_size=patch_size)(img).shape)
    print("RandGridPatch", patch_size, RandGridPatch(patch_size=patch_size)(img).shape)

Current behavior:

GridPatch (0, 4, 4) torch.Size([4, 1, 8, 4, 4])
RandGridPatch (0, 4, 4) ZeroDivisionError: integer modulo by zero
GridPatch (None, 4, 4) torch.Size([4, 1, 8, 4, 4])
RandGridPatch (None, 4, 4) TypeError: unsupported operand type(s) for %: 'int' and 'NoneType'

Expected behavior
RandGridPatch should honor the documented 0 / None whole-dimension behavior and return patches like GridPatch for these inputs.

Environment

MONAI version: 1.6.0rc1+48.g8690ae74
Numpy version: 1.26.4
Pytorch version: 2.10.0+cpu
MONAI rev id: 8690ae74a8a489d31fe1f9ac8ef0bff63165383e
System: Windows-11-10.0.26200-SP0
Python version: 3.12.1

Additional context
The likely failing expression is in monai/transforms/spatial/array.py, where default max_offset is computed as tuple(s % p for s, p in zip(array.shape[1:], self.patch_size)).

Read monai/transforms/spatial/array.py, focusing on RandGridPatch.call and its default max_offset calculation. Re-run the provided 0 and None examples, then verify RandGridPatch and RandGridPatchd produce the documented whole-dimension shapes matching GridPatch.