Raise clear ValueError for empty Dataset in stats()

[brendancol]

Closes #2637

stats() accepts an xarray Dataset and runs per-variable, then merges the results. If the Dataset has no data variables, the merge hit result = dfs[0] on an empty list and raised an opaque IndexError that told the caller nothing.

This adds an early check in the Dataset branch: when values.data_vars is empty, raise a ValueError saying there is nothing to compute statistics over, before reaching the dfs[0] access.

• Guard the empty-Dataset case in the stats() Dataset branch with a clear ValueError.
• Add a regression test that passes an empty Dataset and asserts ValueError.

Backend coverage: the Dataset branch is shared by all backends and the guard runs before any backend dispatch, so numpy / cupy / dask+numpy / dask+cupy are all covered.

Test plan:

• New test test_stats_empty_dataset_raises_value_error_2637 passes
• Existing Dataset return_type test still passes

[brendancol]

PR Review: Raise clear ValueError for empty Dataset in stats()

Blockers (must fix before merge)

None.

Suggestions (should fix, not blocking)

None.

Nits (optional improvements)

• xrspatial/zonal.py:861 -- len(values.data_vars) == 0 works and is clear. if not values.data_vars: would be slightly more idiomatic since the data_vars mapping is falsy when empty, but the explicit length check reads fine and matches the surrounding style. Take it or leave it.

What looks good

• The guard is in the right place: after the return_type check and before the dfs[0] access, so the empty case is caught early with a message that names the actual problem (no data variables).
• The error message tells the caller what to do (pass a Dataset with at least one data variable), not just what went wrong.
• The regression test reuses the existing small_zones_values_2558 fixture and asserts on the "no data variables" text, so the old IndexError would not satisfy it.
• Scope is tight: two files, no unrelated edits.

Checklist

• Algorithm matches reference/paper: n/a (input validation fix)
• All implemented backends produce consistent results: yes, guard runs before backend dispatch
• NaN handling is correct: n/a
• Edge cases are covered by tests: yes, empty Dataset is the case in question
• Dask chunk boundaries handled correctly: n/a
• No premature materialization or unnecessary copies: yes, the data_vars length check is cheap
• Benchmark exists or is not needed: not needed
• README feature matrix updated: not needed, no new public API
• Docstrings present and accurate: stats() docstring unchanged, still accurate