enh: rewrite duckdb groupby so it avoids group-by-all (#3267)

rewrite duckdb groupby so it avoids group-by-all#

Author: MarcoGorelli

CONTRIBUTING.md

@@ -185,9 +185,16 @@ If you add code that should be tested, please add tests.
   - Never assume that your data is ordered in any pre-defined way.
   - Never materialise your data (only exception: `collect`).
   - Avoid calling the schema / column names unnecessarily.
-  - For DuckDB, use the Python API as much as possible, only falling
+
+- DuckDB:
+
+  In addition to the above:
+
+  - Use the Python API as much as possible, only falling
     back to SQL as the last resort for operations not yet supported
     in their Python API (e.g. `over`).
+  - Use standard SQL constructs where possible instead of non-standard
+    ones such as `GROUP BY ALL` or `EXCLUDE`.
 
 ### Test Failure Patterns
 

narwhals/_duckdb/group_by.py

@@ -3,6 +3,7 @@
 from itertools import chain
 from typing import TYPE_CHECKING
 
+from narwhals._duckdb.utils import join_column_names
 from narwhals._sql.group_by import SQLGroupBy
 
 if TYPE_CHECKING:
@@ -27,7 +28,12 @@ def __init__(
         self._compliant_frame = frame.drop_nulls(self._keys) if drop_null_keys else frame
 
     def agg(self, *exprs: DuckDBExpr) -> DuckDBLazyFrame:
-        agg_columns = tuple(chain(self._keys, self._evaluate_exprs(exprs)))
-        return self.compliant._with_native(
-            self.compliant.native.aggregate(agg_columns)  # type: ignore[arg-type]
-        ).rename(dict(zip(self._keys, self._output_key_names)))
+        agg_columns = tuple(self._evaluate_exprs(exprs))
+        result = self.compliant.native.aggregate(
+            tuple(chain(self._keys, agg_columns)),  # type: ignore[arg-type]
+            join_column_names(*self._keys),
+        )
+
+        return self.compliant._with_native(result).rename(
+            dict(zip(self._keys, self._output_key_names))
+        )