Improved role assignment

Current role assignment may not always optimise the best parameters. For example, if you have fewer stateless class processes in your cluster than proxies configured, you will get fewer proxies than you asked for. You may prefer FoundationDB to run proxies on other processes e.g. storage servers. But this needs some thought, as then you may get unpredictable performance as some storage servers will have more CPU load than others.

At the very least I think FoundationDB should warn to the user undesirable configurations of process sharing (particularly log and storage sharing) or if recruited counts are less than desired counts.

Start by tracing FoundationDB's role-assignment and process-recruitment paths, focusing on stateless, proxy, log, and storage processes. Define expected behavior for fewer available processes, process sharing, and recruited counts, then determine how configuration warnings should be reported and validate the behavior across these cases.

Written by the indexing model from the issue text.