[multi stage] Measure threadCpuTime metric (similar to existing engine) in the operators, aggregate and include in response

In the current engine, we measure threadCpuTime metric at the BaseCombineOperator level. Aggregated metric is sent to servers and broker aggregates across servers and includes in query response stats. We need to do something similar (granularity etc to be discussed) in the multi-stage engine.

Start by reading the existing threadCpuTime measurement at BaseCombineOperator and tracing how metrics move through the current engine, servers, broker, and query response stats. Compare those entry points with the multi-stage engine operators; the work is done when the agreed granularity is measured, aggregated across servers, and included in the response statistics.

Written by the indexing model from the issue text.