Fix zero-padding edge case in tile stitching

[Dartayous]

What is the purpose of this PR?

This PR fixes a tile-stitching edge case related to issue #59.

When bottom or right padding is 0, the current _stitch_tiles() slicing logic can collapse the stitched output to an empty spatial dimension because Python evaluates -0 as 0.

Closes #59

How did you implement your changes?

Updated Nimbus._stitch_tiles() in src/nimbus_inference/nimbus.py.

The previous implementation used:

stitched[:, :, padding[0] : -padding[1], padding[2] : -padding[3]]

When padding[1] == 0 or padding[3] == 0, this becomes a start:0 slice and can produce an output shape like (1, 1, 0, 0).

The fix explicitly uses None for zero end-padding:

h_end = None if padding[1] == 0 else -padding[1]
w_end = None if padding[3] == 0 else -padding[3]

stitched = stitched[:, :, padding[0]:h_end, padding[2]:w_end]

I also added a regression test, test_stitch_tiles_handles_zero_end_padding, which verifies that zero bottom/right padding preserves the expected stitched output shape.

Validation:

pytest tests/test_nimbus.py -v
7 passed

Remaining issues

None from my testing. I would appreciate maintainer feedback on whether this is the preferred place to handle the zero-padding crop edge case.

[JLrumberger]

Thanks @Dartayous, the code looks good.