Toward Attention-based TinyML: A Heterogeneous Accelerated Architecture and Automated Deployment Flow (Full Report)