add long running network tests that test different workloads

This means for example adding and catting different things, of different sizes.

This could use docker.

This could allocate a couple of machines to it, for 10-20min, and we could have travis machines start up + participate and shut down, etc.

Scripts could randomly add + pull things over time, simulating use. Scripts could go through heavy spikes of traffic, etc.

Long running tests would have to report stats to a master process that tracks + logs everything and that can be explored over time.

Rather than a test "succeeding" or "failing" we would see metrics like "avg time to completing a block download" or network bandwidth breakdowns (by subprotocol, etc).