scale the socket receive buffer on the fly #39092

As an experiment, I tried scaling the receive buffer on the fly depending on how full it was on the last iteration It has the advantage of using less memory for slow streams and being more efficient for faster streams

It does not allow for manual control (but this can probably be added) and it effectively short-circuits the back pressure mechanism - as the first read always gets through - which is not very well suited for stream originating from the network