[Tensorpipe Agent] Tracking Active Call Metrics (#38265)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38265
Tracking the active calls counts in the TensorPipe Agent:
* clientActiveCalls: running count of un-responded/un-errored RPC's sent
* serverActiveCalls: running count of un-responded RPC's received
* serverAsyncCallCount: running count of received RPC's set to be completed asynchronously
ghstack-source-id: 103949984
Test Plan: CI
Reviewed By: lw
Differential Revision: D21508957
fbshipit-source-id: 8be9dbf77ec06c138c8dd70443976d7bccee0f1e