Draft: Multi-role benchmarks running across multi-node SUTs

This reworks the plan schema (in a back-compat manner) and the resultstore schama (in non-back-compat manner) to allow definition of benchmarks with multiple, concurrent roles, that can be mapped across multiple nodes. A SUT may now be composed of more than 1 node.

Earlier commits prepare the ground by refactoring and otherwise improving the code. documentation is updated and "plan exec" is finally reworked to support multi-role benchmarks. result analysis is not yet updated for the new schama and is likely broken. I wanted to get the ball rolling on review before waiting for that though.