Temporary fix for filtering on empty batches (#1901)

Potential fix for #1804

Might want to write a test, but not sure yet how to reproduce without
using glue.

Closes #1804

Author: koenvo

pyiceberg/io/pyarrow.py

@@ -1421,11 +1421,15 @@ def _task_to_record_batches(
 
             # Apply the user filter
             if pyarrow_filter is not None:
-                current_batch = current_batch.filter(pyarrow_filter)
+                # Temporary fix until PyArrow 21 is released ( https://github.com/apache/arrow/pull/46057 )
+                table = pa.Table.from_batches([current_batch])
+                table = table.filter(pyarrow_filter)
                 # skip empty batches
-                if current_batch.num_rows == 0:
+                if table.num_rows == 0:
                     continue
 
+                current_batch = table.combine_chunks().to_batches()[0]
+
             result_batch = _to_requested_schema(
                 projected_schema,
                 file_project_schema,