bump: `deepspeed` to be 0.16 (#21194)
* bump: `deepspeed` to be 0.16
* feat: enforce DeepSpeed version requirement for PyTorch >= 2.6 and update checkpoint loading behavior
---------
Co-authored-by: Justus Schock <12886177+justusschock@users.noreply.github.com>
Co-authored-by: Deependu Jha <deependujha21@gmail.com>