add available memory check to accelerators (#4508)
* add available memory check to accelerator
* catch case where nvmlInit fails
* add pynvml to reqs
* fix for cpu systems
* Update accelerator/cuda_accelerator.py
Co-authored-by: Michael Wyatt <michaelwyatt@microsoft.com>
* simplify
---------
Co-authored-by: Michael Wyatt <michaelwyatt@microsoft.com>
