Unified Modeling of Multi-Talker Overlapped Speech Recognition and Diarization with a Sidecar Separator