GitLab

* Remove unused headers
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Unify the fused attn workspace size cpp code
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Reduce the skipped cases
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Rename self/cross attention to qkvpacked/kvpacked
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Update attention mask docs
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Refine the attn mask implementations
Signed-off-by: Reese Wang <rewang@nvidia.com>

---------
Signed-off-by: Reese Wang <rewang@nvidia.com>