[Don-Yin]

A bucket containing [NaN, -7, 3] currently returns NaN from get_max and get_abs_max with skipna=True, while get_min(..., skipna=False) returns -7 instead of propagating the NaN.

This keeps NaNs behind valid values in the descending sort and reuses the existing missing-bin mask for skipna=False. Empty and all-NaN buckets keep their current behavior; no fill-value semantics are changed.

The new cases cover all three methods, both skipna settings, three chunk sizes, Dask arrays and xarray wrappers, and verify that constructing the result stays lazy. They also check valid, empty, all-NaN, and out-of-area samples.

Each row contains six measured cases: three chunk sizes × two input wrappers. On the unchanged base, 18 of the 36 new cases fail; with the fix, all 36 pass.

Tests run:

• pytest -q pyresample/test/test_bucket.py: 77 passed.

• pre-commit run --files pyresample/bucket/__init__.py pyresample/test/test_bucket.py: Ruff, whitespace checks, Bandit, mypy and isort passed.

• git diff --check: passed.

• Tests added

• Tests passed

• Fully documented: restores the existing documented skipna behavior.