Add docs about how to use Karpenter to scale to 0 and launch new nodes on demand

[spwoodcock]

• We have dynamic determination of CPUs / RAM for processing, depending on imagery size.
• Karpenter can be used to keep resources small, but then spawn appropriate sized worker nodes on demand, depending on RAM requirements.
• This approach is used currently at HOT & should be documented, to help those deploying in AWS.

[linear]

TECH-1173