refactor(duckdb): simplify to_pyarrow_batches implementation

ibis/backends/duckdb/__init__.py

@@ -6,6 +6,7 @@
 import itertools
 import os
 import warnings
+import weakref
 from functools import partial
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Iterable, Iterator, Mapping, MutableMapping
@@ -555,9 +556,6 @@ def to_pyarrow_batches(
             !!! warning "DuckDB returns 1024 size batches regardless of what argument is passed."
         """
         self._import_pyarrow()
-
-        from ibis.backends.duckdb.pyarrow import IbisRecordBatchReader
-
         self._register_in_memory_tables(expr)
         query_ast = self.compiler.to_ast_ensure_limit(expr, limit, params=params)
         sql = query_ast.compile()
@@ -574,7 +572,12 @@ def to_pyarrow_batches(
         cursor = con.execute(sql)
 
         reader = cursor.cursor.fetch_record_batch(chunk_size=chunk_size)
-        return IbisRecordBatchReader(reader, cursor)
+        # Use a weakref finalizer to keep the cursor alive until the record
+        # batch reader is garbage collected. It would be nicer if we could make
+        # the cursor cleanup happen when the reader is closed, but that's not
+        # currently possible with pyarrow.
+        weakref.finalize(reader, lambda: cursor.close())
+        return reader
 
     def to_pyarrow(
         self,