Optimize frame body extraction to reduce memory copies

Replace BytesIO.read() with direct buffer slicing to eliminate one
intermediate bytes allocation per received message frame.

Changes:
- Use getbuffer() to get memoryview of underlying buffer
- Slice directly at [body_offset:end_pos] instead of seek+read
- Convert memoryview slice to bytes in single operation
- Maintain buffer position tracking for proper reset behavior

Benefits:
- Eliminates one full-frame allocation on hot receive path
- Particularly impactful for large frames (up to 256MB)
- Maintains compatibility with existing protocol decoder

The memoryview is immediately converted to bytes and released,
preventing buffer resize issues while still gaining the
allocation savings.

Signed-off-by: Yaniv Kaul <ykaul@scylladb.com>

Author: mykaul

cassandra/connection.py

@@ -1260,8 +1260,12 @@ def process_io_buffer(self):
                 return
             else:
                 frame = self._current_frame
-                self._io_buffer.cql_frame_buffer.seek(frame.body_offset)
-                msg = self._io_buffer.cql_frame_buffer.read(frame.end_pos - frame.body_offset)
+                # Use memoryview to avoid intermediate allocation, then convert to bytes
+                # Explicitly scope the buffer to ensure memoryview is released before reset
+                cql_buf = self._io_buffer.cql_frame_buffer
+                msg = bytes(cql_buf.getbuffer()[frame.body_offset:frame.end_pos])
+                # Advance buffer position to end of frame before reset
+                cql_buf.seek(frame.end_pos)
                 self.process_msg(frame, msg)
                 self._io_buffer.reset_cql_frame_buffer()
                 self._current_frame = None