slope: add degenerate-shape test coverage (1x1, 1xN, Nx1)

[brendancol]

Closes #2697

Adds tests for degenerate raster shapes where a dimension is too small for the 3x3 kernel to have an interior. These shapes already work (output keeps the input shape and comes back all-NaN), but nothing pinned that down. A future change to the kernel bounds could start raising or returning the wrong shape and no test would catch it.

• New test_degenerate_shape_* tests covering shapes (1, 1), (1, 5), (5, 1)
• Planar method on all four backends (numpy, cupy, dask+numpy, dask+cupy), geodesic on numpy
• Test-only change, no source modifications

Backend coverage: numpy, cupy, dask+numpy, dask+cupy.

Test plan:

• pytest xrspatial/tests/test_slope.py -k degenerate -> 15 passed (cupy and dask+cupy ran on a CUDA host, not skipped)
• pytest xrspatial/tests/test_slope.py xrspatial/tests/test_geodesic_slope.py -> 77 passed

[brendancol]

PR Review: slope degenerate-shape test coverage

Blockers

None.

Suggestions

None.

Nits

• test_slope.py:215 The geodesic test asserts shape and all-NaN directly but skips general_output_checks, which test_degenerate_shape_numpy (line 165) calls. The cross-backend helpers already cover attrs/coords on the dask and cupy paths, so this is minor. Adding general_output_checks(raster, result) to the geodesic test would make the attrs/coords/dims check uniform across all five tests.

What looks good

• All four planar backends plus geodesic are covered through the existing helpers in general_checks.py.
• nan_edges=False is correct here. These outputs are entirely NaN, so assert_nan_edges_effect would have no non-NaN interior to compare against.
• The comment explains why these shapes come back all-NaN (the range(1, rows-1) loop is empty when rows<=2), so the test pins a real contract instead of freezing whatever the code happens to return today.
• Ran locally on a CUDA host: 15 degenerate tests passed, with cupy and dask+cupy actually executing rather than skipping.

Checklist

• Test-only, no algorithm change
• All implemented backends produce consistent results
• NaN handling is correct
• Edge cases covered
• Dask chunk boundaries handled correctly
• No premature materialization
• Benchmark not needed; README matrix and docstrings not applicable

[brendancol]

Follow-up review

The one nit from the first pass is resolved: the geodesic degenerate-shape test now calls general_output_checks(raster, result), so attrs/coords/dims preservation is asserted uniformly across all five degenerate tests (952062e).

No remaining Blockers, Suggestions, or Nits. Re-ran pytest xrspatial/tests/test_slope.py -k degenerate -> all pass.