Fix Coordinates.to_index performance regression

The codes passed to pd.MultiIndex were being converted from cache-friendly
ndarrays into Python lists to silence a mypy arg-type error introduced in
pydata#10694. The extra per-element conversion dominates runtime for large indexes
(~13s on a 100x2000x300 array). Pass the ndarrays directly and suppress the
type error the same way as for `levels` just above.

Fixes pydata#11305

Co-authored-by: Claude <noreply@anthropic.com>

Author: thodson-usgs

doc/whats-new.rst

@@ -26,6 +26,10 @@ Deprecations
 Bug Fixes
 ~~~~~~~~~
 
+- Fix a major performance regression in :py:meth:`Coordinates.to_index` (and
+  consequently :py:meth:`Dataset.to_dataframe`) caused by converting the cached
+  code ndarrays into Python lists (:issue:`11305`).
+
 
 Documentation
 ~~~~~~~~~~~~~

xarray/core/coordinates.py

@@ -194,7 +194,7 @@ def to_index(self, ordered_dims: Sequence[Hashable] | None = None) -> pd.Index:
 
         return pd.MultiIndex(
             levels=level_list,  # type: ignore[arg-type,unused-ignore]
-            codes=[list(c) for c in code_list],
+            codes=code_list,  # type: ignore[arg-type,unused-ignore]
             names=names,
         )