Update how process groups form with TCPStore instead

[py-rh]

When multiple different requests are made to the same putter, using global env var to set up multiple process groups is bad and leads to one of them trying to work on the wrong port and hanging. TCPStore makes it explicit. (Factored out the setup in multiple methods)

But we still might have multiple requests trying and spin up multiple process groups at the same time. At the cost of worse performance, the simple approach was adding a semaphore. The right approach might be doing work in a process pool and isolating them.