controller: detect cyclical resource dependencies

It is possible for the resource controller to wait forever. At some point in the lifecycle, walk the dependency graph and raise an error if cycles are detected.

Start with the resource controller lifecycle and the dependency graph representation. Trace where the controller can wait indefinitely, then determine the appropriate lifecycle point for checking cycles and how errors are reported. Done means cyclical dependencies are detected and the controller raises an error instead of waiting forever.

Written by the indexing model from the issue text.