Cluster refactor

[approxit]

Warning, massive changes ahead!

What I've done:

• Made business logic object-centric instead of service-centric
• Each object is responsible for itself as much as possible
• Support for multiple clusters at the same time
• Most of the client requests are pointing to a specific cluster
• RayService handles only ray-related requests and cluster collection
• GolemService has become quite small due to logic being moved elsewhere
• Cluster is responsible for the nodes collection and manager stack collection
• ClusterNode is responsible for its state, data, activity, and sidecar tasks collection
• ClusterNodeSidecar takes responsibility for dedicated logic that is run alongside the cluster node

Notable remarks:

• Removal of all nodes in the cluster currently does nothing special - cluster will still exist and manager stacks will still run.
• Successful head node start could stop its manager stack
• New versions of the services are temporary to still have originals for reference. They will be removed before merging.