[SPARK-57679][PYTHON] Fix numpy type checking

### What changes were proposed in this pull request?

Add a branch to detect `NDArray` for `numpy >= 2.5.0`.

### Why are the changes needed?

Our detection for `NDArray` is a bit fragile. `numpy` changed the type alias in `2.5.0` so we need another way to detect it.

CI is failing - https://github.com/apache/spark/actions/runs/28064134730/job/83087583742

### Does this PR introduce _any_ user-facing change?

No.

### How was this patch tested?

Local test passed

### Was this patch authored or co-authored using generative AI tooling?

Yes, Claude Code (Opus 4.8 high)

Closes #56757 from gaogaotiantian/fix-numpy-type-checking.

Authored-by: Tian Gao <gaogaotiantian@hotmail.com>
Signed-off-by: Tian Gao <gaogaotiantian@hotmail.com>
(cherry picked from commit 5f7570c)
Signed-off-by: Tian Gao <gaogaotiantian@hotmail.com>

Author: gaogaotiantian

python/pyspark/pandas/typedef/typehints.py

@@ -155,8 +155,18 @@ def as_spark_type(
         and hasattr(tpe, "__args__")
         and len(tpe.__args__) > 1
     ):
-        # numpy.typing.NDArray
+        # numpy.typing.NDArray for numpy < 2.5
         return types.ArrayType(as_spark_type(tpe.__args__[1].__args__[0], raise_error=raise_error))
+    elif (
+        hasattr(tpe, "__origin__")
+        and hasattr(tpe.__origin__, "__value__")
+        and getattr(tpe.__origin__.__value__, "__origin__", None) is np.ndarray
+        and hasattr(tpe, "__args__")
+        and len(tpe.__args__) > 0
+    ):
+        # numpy.typing.NDArray for numpy >= 2.5: a PEP 695 type alias whose __value__
+        # resolves to np.ndarray[shape, dtype[scalar]], with the scalar at __args__[0]
+        return types.ArrayType(as_spark_type(tpe.__args__[0], raise_error=raise_error))
 
     if isinstance(tpe, np.dtype) and tpe == np.dtype("object"):
         pass