ensure that subscription connections are QUIT cleanly (interactive connections were already QUIT), *without* resurrecting like zombies

We're stable enough to not need this anymore, yay for faster runs!

As we go for green on the test suite, we'll want this checked in PRs and such. And in general DEBUG and RELEASE not matching is becoming a pain. Trying to minimize the differences.

This moves the socket bits all the physical connection which I think is a lot easier to grok than the double relationship between it and SocketManager. Pushing a PR for sanity check.

This is pretty much a boolean now...let's make it one. Also note PhysicalConnection is passes as `this`, it can't be null.

This helps for real-time output on long-running or hanging tests.

make sure there is only **one** path that can cause endpoints to be added, and ensure that the activation happens after the constructor - removes race condition that can lead to abandoned endpoints; note as part of this, switch the server-snapshot to span-based

This fixes net462, but not netcoreapp test hangs

Marc's working on connection code so we'll plug these in after.

This simplifies internal logistics and our profiling a bit

Fix #838 - provide mechanism for app-code to compute hash-slot using the muxer's state; also avoid hash-slot overhead when not cluster