docs: Correct documentation to state that NVIDIA container runtime is included, not CUDA toolkit

[bryantbiggs]

Issue number:

Closes #

Description of changes:

• Correct documentation to state that NVIDIA container runtime is included, not the CUDA toolkit
  • The NVIDIA container runtime (usually installed/configured via the NVIDIA *container toolkit) is required on the host when wanting to utilize NVIDIA GPUs within your containers
  • The NVIDIA CUDA toolkit is a user space component for developing CUDA based applications - this not required on the host, nor should it be for containerized environments

XRef - awslabs/amazon-eks-ami#1924 (comment)

Testing done:
None

Terms of contribution:

By submitting this pull request, I agree that this contribution is dual-licensed under the terms of both the Apache License, version 2.0, and the MIT license.

[bcressey]

LGTM. It is indeed just the subset of drivers that the NVIDIA container runtime expects which are packaged on the host.

I'd also drop the "included in your orchestrated containers" bit, since it doesn't make sense now.

The previous documentation was essentially written from the container's perspective - "if you have a container that includes the CUDA toolkit, then the nvidia variants have all the libraries etc needed for your container to work."