Automated capacity management

Due to asynchronous nature Engine can accept for execution more plans than it is physically capable of executing. A partial answer to this problem is to use specify plan concurrency and apply back pressure to incoming plans. The problems with this approach are:
- inefficiency in situations where different plan types have very different resource utilizations
- difficult to figure out correct concurrency level
- best concurrency level may change over time

This task is to investigate automatic capacity management. Some ideas:
- in real time, measure utilization of key resources and and distribute cost over plan types being executed
- estimate load based on sample of traces of executed plans and correlate it with resource utilization

No files, tests, or entry points are named. Start by reviewing the existing plan-concurrency and back-pressure behavior, then investigate how execution traces and key-resource utilization could be measured and related to plan types. Done means producing a concrete automatic capacity-management approach that addresses changing workloads and differing resource use.

Written by the indexing model from the issue text.