Fix mypy errors with numpy 2.3 type stubs (#10792)

NumPy 2.3 has stricter type annotations that revealed legitimate type
issues in our codebase. This commit:

1. Updates numpy constraint: removes numpy<2.3 and updates numba>=0.62
   since numba 0.62+ supports numpy 2.3
2. Fixes type annotations in indexing.py for lists that hold mixed types
3. Adds appropriate type ignores for known limitations

Fixes #10791

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-authored-by: Claude <[email protected]>

Author: max-sixty

pyproject.toml

@@ -30,10 +30,9 @@ accel = [
   "scipy>=1.13",
   "bottleneck",
   "numbagg>=0.8",
-  "numba>=0.59",
+  "numba>=0.62",  # numba 0.62 added support for numpy 2.3
   "flox>=0.9",
   "opt_einsum",
-  "numpy<2.3",    # numba has not updated yet: https://github.com/numba/numba/issues/10105
 ]
 complete = ["xarray[accel,etc,io,parallel,viz]"]
 io = [

xarray/core/indexing.py

@@ -694,8 +694,8 @@ def __init__(self, array: Any, key: ExplicitIndexer | None = None):
         """
         if isinstance(array, type(self)) and key is None:
             # unwrap
-            key = array.key  # type: ignore[has-type]
-            array = array.array  # type: ignore[has-type]
+            key = array.key  # type: ignore[has-type, unused-ignore]
+            array = array.array  # type: ignore[has-type, unused-ignore]
 
         if key is None:
             key = BasicIndexer((slice(None),) * array.ndim)
@@ -1251,8 +1251,8 @@ def _decompose_vectorized_indexer(
     if indexing_support is IndexingSupport.VECTORIZED:
         return indexer, BasicIndexer(())
 
-    backend_indexer_elems = []
-    np_indexer_elems = []
+    backend_indexer_elems: list[slice | np.ndarray[Any, np.dtype[np.generic]]] = []
+    np_indexer_elems: list[slice | np.ndarray[Any, np.dtype[np.generic]]] = []
     # convert negative indices
     indexer_elems = [
         np.where(k < 0, k + s, k) if isinstance(k, np.ndarray) else k

xarray/tests/test_dataarray.py

@@ -3883,7 +3883,7 @@ def test_to_masked_array(self) -> None:
         v = range(N)
         da = DataArray(v)
         ma = da.to_masked_array()
-        assert len(ma.mask) == N
+        assert isinstance(ma.mask, np.ndarray) and len(ma.mask) == N
 
     def test_to_dataset_whole(self) -> None:
         unnamed = DataArray([1, 2], dims="x")

xarray/tests/test_namedarray.py

@@ -374,7 +374,7 @@ def test_duck_array_class(self) -> None:
 
         masked_a: np.ma.MaskedArray[Any, np.dtype[np.int64]]
         masked_a = np.ma.asarray([2.1, 4], dtype=np.dtype(np.int64))  # type: ignore[no-untyped-call]
-        check_duck_array_typevar(masked_a)
+        check_duck_array_typevar(masked_a)  # type: ignore[arg-type]  # MaskedArray not in duckarray union
 
         custom_a: CustomArrayIndexable[Any, np.dtype[np.int64]]
         custom_a = CustomArrayIndexable(numpy_a)

xarray/tests/test_parallelcompat.py

@@ -90,7 +90,7 @@ def from_array(
     def rechunk(self, data: DummyChunkedArray, chunks, **kwargs) -> DummyChunkedArray:
         return data.rechunk(chunks, **kwargs)
 
-    def compute(self, *data: DummyChunkedArray, **kwargs) -> tuple[np.ndarray, ...]:
+    def compute(self, *data: DummyChunkedArray, **kwargs) -> tuple[np.ndarray, ...]:  # type: ignore[override]
         from dask.array import compute
 
         return compute(*data, **kwargs)