Framework for scheduling feature-compatible Contracts & Nodes

This idea has come up in two contexts:

1. maximum request size to compute node
2. external libraries loaded by compute node (e.g. MXNet)

Based on the discussion so far, the proposal is:

• allow the contract to register a list of required resources (e.g., libraries, max request size)
• allow the compute nodes to maintain a /etc/ekiden/compute.conf (just like a webserver)
• have the scheduler assign contracts to nodes which provide a superset of the required resources
• bonus: compute nodes can charge for providing more resources

Security note: The types of resource requests need to be kept generic enough so that a malicious contract can't single out a particular group of compute nodes. Perhaps even not scheduling contracts that don't hit a target percentage of all compute nodes.

Start by reviewing the scheduler, contract resource requirements, and compute-node configuration described in the issue. Trace how contracts are assigned to nodes and how nodes could expose resources through /etc/ekiden/compute.conf. Done means a decided, secure compatibility model covering resource registration, superset matching, and the privacy concern around identifying node groups.