End-to-End Transformer Acceleration Through Processing-in-Memory Architectures