MultiHeadAttention op shall return attention probabilities

[amancini-N]

Support pointer-generator networks in MultiHeadAttention op. Specifically, MultiHeadAttention op shall have an additional output returning the attention probabilities (softmax result) out.

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.