Cache the result of DaskManager.normalize_chunks

This is only used with the backends codepath, where the inputs
are guaranteed to be tuples. By contrast, `dask.array.normalize_chunks`
accepts dicts as inputs and so, is harder to cache transparently

xarray/namedarray/daskmanager.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 from collections.abc import Callable, Iterable, Sequence
+from functools import lru_cache
 from typing import TYPE_CHECKING, Any
 
 import numpy as np
@@ -19,12 +20,20 @@
 
     try:
         from dask.array import Array as DaskArray
+
     except ImportError:
         DaskArray = np.ndarray[Any, Any]
 
 
 dask_available = module_available("dask")
 
+if dask_available:
+    from dask.array.core import normalize_chunks
+
+    normalize_chunks = lru_cache(normalize_chunks)
+else:
+    normalize_chunks = None
+
 
 class DaskManager(ChunkManagerEntrypoint["DaskArray"]):
     array_cls: type[DaskArray]
@@ -52,8 +61,6 @@ def normalize_chunks(
         previous_chunks: _NormalizedChunks | None = None,
     ) -> Any:
         """Called by open_dataset"""
-        from dask.array.core import normalize_chunks
-
         return normalize_chunks(
             chunks,
             shape=shape,