Improve cost optimization and documentation

[jgbradley1]

• Add a new k8s node pool dedicated to indexing jobs (with high memory VMs) with autoscaling enabled and size=0 by default to reduce cost when no indexing jobs are active.
• Change VM type of graphrag api server node pool to leverage cheaper VMs that are simultaneously sufficient for query compute requirements.
• Update architecture diagram to represent the new simplified networking architecture