perf: vectorize is_regular_1d for ndarray inputs

Per-element Python iteration over a 100K-element int64 array dominated
create_array runtime on the (10**8,) chunks=(1000,) regression case
(~6 ms in is_regular_nd, downstream of ChunksTuple normalization).
Dispatch on np.ndarray and use a single vectorized comparison instead.

End-to-end create_array on the regression case: ~7.9 ms -> ~0.6 ms.

Promote numpy to a runtime import (was TYPE_CHECKING-only) for the
isinstance dispatch.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: d-v-b

src/zarr/core/metadata/v3.py

@@ -5,6 +5,7 @@
 from dataclasses import dataclass, field, replace
 from typing import TYPE_CHECKING, Any, Final, Literal, NotRequired, TypeGuard, cast
 
+import numpy as np
 from typing_extensions import TypedDict
 
 from zarr.abc.codec import ArrayArrayCodec, ArrayBytesCodec, BytesBytesCodec, Codec
@@ -39,8 +40,6 @@
 if TYPE_CHECKING:
     from typing import Self
 
-    import numpy as np
-
     from zarr.core.buffer import Buffer, BufferPrototype
     from zarr.core.chunk_grids import ChunksTuple
     from zarr.core.dtype.wrapper import TBaseDType, TBaseScalar
@@ -385,6 +384,9 @@ def is_regular_1d(
     if len(dim_chunks) <= 1:
         return True
     first = dim_chunks[0]
+    if isinstance(dim_chunks, np.ndarray):
+        # Vectorized comparison avoids per-element Python iteration over int64 arrays.
+        return bool((dim_chunks[1:-1] == first).all() and dim_chunks[-1] <= first)
     for c in dim_chunks[1:-1]:
         if c != first:
             return False