fix: Filter sorted flag from physical in CategoricalChunked (#23827)

Author: orlp

crates/polars-core/src/chunked_array/logical/categorical.rs

@@ -35,7 +35,13 @@ impl<T: PolarsCategoricalType> CategoricalChunked<T> {
     }
 
     pub(crate) fn get_flags(&self) -> StatisticsFlags {
-        self.phys.get_flags()
+        // If we use lexical ordering then physical sortedness does not imply
+        // our sortedness.
+        let mut flags = self.phys.get_flags();
+        if self.uses_lexical_ordering() {
+            flags.set_sorted(IsSorted::Not);
+        }
+        flags
     }
 
     /// Set flags for the ChunkedArray.

py-polars/tests/unit/operations/namespaces/test_categorical.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+from io import BytesIO
+
 import pytest
 
 import polars as pl
@@ -236,3 +238,17 @@ def test_cat_slice() -> None:
         "",
         None,
     ]
+
+
+def test_cat_order_flag_csv_read_23823() -> None:
+    data = BytesIO(b"colx,coly\nabc,123\n#not_a_row\nxyz,456")
+    lf = pl.scan_csv(
+        source=data,
+        comment_prefix="#",
+        schema_overrides={"colx": pl.Categorical},
+    )
+    expected = pl.DataFrame(
+        {"colx": ["abc", "xyz"], "coly": [123, 456]},
+        schema_overrides={"colx": pl.Categorical},
+    )
+    assert_frame_equal(expected, lf.sort("colx", descending=False).collect())