Implement Runner job proxying

This is the main implementation issue of the epic. I'm planning to create MRs as needed, implementing individual bits of work in chunks that make sense (and are easier to review).

Runner (un)registers with kas by making an RPC call. Runner supplies information about itself, including the authentication token, as part of the call. kas stores the information in Redis (except for the token!) so that it can be later looked up based on capability hash, etc. Details to be worked out as part of the issue.