Elastic training support (deepspeedai#602) Co-authored-by: Samyam Rajbhandari <samyamr@microsoft.com>

bump to 0.3.8 

calculate grad norm wrt sub partitions 

bump to 0.3.7 

bump to 0.3.6 and fix manifest to include reqs (deepspeedai#561) 

bump to 0.3.5 

bump version 0.3.4 

bump to v0.3.3 

bump version 

DeepSpeed JIT op + PyPI support (deepspeedai#496) Co-authored-by: Shaden Smith <Shaden.Smith@microsoft.com> Co-authored-by: Reza Yazdani <reyazda@microsoft.com>