Add Dual Chunk Attention (DCA) for long-context training#4048

Ternura143 wants to merge 5 commits intoNVIDIA:mainNVIDIA/Megatron-LM:mainfrom

Ternura143:feature/dual-chunk-attentionTernura143/Megatron-LM:feature/dual-chunk-attentionCopy head branch name to clipboard