Lean Attention: Hardware-Aware Scalable Attention Mechanism for the Decode-Phase of Transformers