Fix incorrect connection metrics tracking on instances pending shutdown

[bdach]

While a spectator server instance was pending shutdown, it could start reporting endlessly rising user counts until its actual shutdown. The reason for this was questionable logic in ConcurrentConnectionLimiter.

When I wrote that thing I used try...finally in OnConnectedAsync() and OnDisconnectedAsync() because I wanted to have multiple returns in the body of the method and sitll call await next() at the end. However I appear to have also forgotten that it means that await next() would be called even if the preceding code threw an exception, and then the exception would be thrown anyway.

This in the case of StatefulUserHubs led to the following sequence of events:

1. ConcurrentConnectionLimiter.OnConnectedAsync() fires
2. It dies on its innards, but because of the finally, calls await next() anyway
3. Inside await next() is the hub's connection method, in particular LoggingHub.OnConnectedAsync(), which increments the datadog connected user gauge
4. Then after the finally, because any exception that was thrown was not caught, it will get rethrown, interrupting the connection flow
5. After this, OnDisconnectedAsync() will not get called for this connection, leading the counter to not decrement back down from the increment in point (3)

leading to the runaway increase in reported user counts.

To fix, stop using try...finally and split the blocks of code that want early-returns to separate methods instead.