Add Examples sections and fix cross-correlation wording in edge_detection docstrings

[brendancol]

Closes #3674

• Adds an Examples section to each of the five public functions (sobel_x, sobel_y, prewitt_x, prewitt_y, laplacian). The module had no >>> blocks at all; every example here was executed and is doctest-verified.
• Rewords the four gradient docstrings to say the kernel is applied by cross-correlation (scipy.ndimage.correlate semantics) and states the sign convention. The old "convolving with" wording predicted the negated result for the antisymmetric Sobel/Prewitt kernels: on a left-to-right ramp, sobel_x returns +8 where convolving the documented kernel gives -8. The code matches the standard Sobel/Prewitt convention, so only the wording changes.
• Adds drift-guard tests modeled on test_docstring_params_match_signature and test_docstring_states_all_backends: documented parameters must match the signatures, Examples sections must exist, and the cross-correlation wording is pinned.

Doc-only; no behavior changes. Also updates the documentation sweep state CSV.

Overlap note: #3678 (error-handling sweep) documents NaN propagation in these same docstrings via Notes sections. I dropped my NaN wording to stay out of its way, but both PRs insert text right after Returns, so whichever merges second will need a trivial conflict resolution (numpydoc order puts Notes before Examples).

Test plan:

• python -m doctest on the module: 31 examples pass
• pytest xrspatial/tests/test_edge_detection.py: 101 passed
• flake8: no new findings (the 4 E201s in the kernel literals and the E302 in the test file predate this PR)

[brendancol]

PR Review: Add Examples sections and fix cross-correlation wording in edge_detection docstrings

Blockers (must fix before merge)

None.

Suggestions (should fix, not blocking)

• Each gradient docstring now says "cross-correlating" twice: once in the lead sentence ("Detects vertical edges by cross-correlating with...") and again in the paragraph right after the kernel ("The kernel is applied by cross-correlation, as in scipy.ndimage.correlate..."), e.g. xrspatial/edge_detection.py:34 and :39. The second paragraph earns its keep by naming the scipy equivalent and the sign convention, but the two could be merged into one sentence if you want the docstrings tighter.
• Coordination, not a defect: #3678 inserts Notes sections into these same five docstrings right after Returns, where this PR inserts Examples. Whichever merges second gets a conflict at every insertion point. The PR body already says this; resolution is mechanical (numpydoc order puts Notes before Examples).

Nits (optional improvements)

• laplacian's docstring skips the cross-correlation paragraph. Correct, since the kernel is symmetric and the two operations agree, but a reader comparing the five docstrings may wonder why it's absent. A short clause ("the kernel is symmetric, so correlation and convolution agree") would preempt that. Fine to leave as is.

What looks good

• All five examples are real output: they pass bare doctest.testmod (31 examples) and pytest --doctest-modules on this host.
• The sign-convention fix is backed by measurement, not reading: sobel_x on a left-to-right ramp returns +8, matching scipy.ndimage.correlate, while convolving the documented kernel gives -8. The old wording predicted the wrong sign.
• The drift-guard tests follow the existing patterns (test_docstring_params_match_signature in test_multispectral.py, phrase-pinning as in test_proximity.py) and pin "cross-correlat" / reject "by convolving" so the wording can't silently regress.
• Doc-only: no function body changed, and the full test file passes (101 tests). Flake8 findings are unchanged from baseline (4 E201 in the kernel literals, 1 E302 in the test file, all pre-existing).

Checklist

• Algorithm matches reference/paper (wording now matches the measured correlate semantics)
• All implemented backends produce consistent results (verified numpy/cupy/dask+numpy/dask+cupy during the audit; unchanged by this PR)
• NaN handling is correct (unchanged; NaN docs land in #3678)
• Edge cases are covered by tests (unchanged coverage plus new docstring-contract tests)
• Dask chunk boundaries handled correctly (n/a, doc-only)
• No premature materialization or unnecessary copies (n/a, doc-only)
• Benchmark exists or is not needed (not needed, doc-only)
• README feature matrix updated (n/a, no new functions or backend changes)
• Docstrings present and accurate

[brendancol]

PR Review follow-up (after 90592c2)

Disposition of the first review's findings:

• Duplicated "cross-correlating"/"cross-correlation" wording: fixed. The paragraph after each gradient kernel now reads "This matches scipy.ndimage.correlate: the response is positive where values increase toward higher column/row index", so the operation is named once, in the lead sentence.
• Overlap with #3678: no change in this PR, by design. Whichever PR merges second resolves the insertion-point conflict; numpydoc order puts its Notes above these Examples.
• laplacian symmetry nit: fixed. Its docstring now says "The kernel is symmetric, so cross-correlation and convolution agree."

Re-verified after the commit: pytest --doctest-modules on the module plus the test file passes (106 tests, includes all 31 docstring examples), and flake8 shows only the pre-existing baseline findings. No further issues.