ENH: Support DataTree in apply_ufunc

[QinXi-ai]

Closes #9789

Problem

apply_ufunc currently treats DataTree as a dict-like object. Unary calls silently return only the root Dataset, while core-dimension and multi-tree calls can fail when child nodes reach the variable dispatch path.

Solution

• Dispatch DataTree inputs before the dict-like Dataset path.
• Reuse map_over_datasets to apply the existing apply_ufunc machinery at corresponding nodes, preserving tree structure, names, inherited coordinates, scalar arguments, kwargs, core dimensions, and tuple outputs.
• Enable the public xarray.ufuncs wrappers to operate over complete trees.
• Add regressions for identity, inherited coordinates, core dimensions plus kwargs/scalars, two isomorphic trees, non-isomorphic errors, multiple outputs, and the ufunc wrapper.

Tests

• python -m pytest -o addopts='' --basetemp D:/CodexCaches/pytest-xarray-9789-full xarray/tests/test_computation.py xarray/tests/test_ufuncs.py xarray/tests/test_datatree_mapping.py -q (323 passed, 131 skipped, 1 xfailed, 2 xpassed)
• ruff check xarray/computation/apply_ufunc.py xarray/ufuncs.py xarray/tests/test_computation.py xarray/tests/test_ufuncs.py
• ruff format --check xarray/computation/apply_ufunc.py xarray/ufuncs.py xarray/tests/test_computation.py xarray/tests/test_ufuncs.py
• git diff --check