Handling Active Call Count through Future Callback (#38589)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38589
This PR creates a unified way of decrementing the active call count on the client side by attaching a callback to the future returned by `TensorPipeAgent::send`.
ghstack-source-id: 104227074
Test Plan: CI/Sandcastle once tests PR's are merged.
Differential Revision: D21605779
fbshipit-source-id: c82396de6984876b09ee032ab1aa0f68a87005be
