Multi-threaded runner

[michielbdejong]

There's no good way to keep track of the message queue in the cluster.
I'm thinking therefore:

• from the primary, start a number of probes, start with just one at at time
• wait for the message traffic in the primary to stop (10ms passing without any messages being sent)
• if that works, increase the number of concurrent probes, for instance one or one in each worker
• there will immediately be a risk of probes competing over the same loops, but that's reality

[michielbdejong]

I had a multi-threaded runner in https://github.com/ledgerloops/jerboa/blob/96776daa05112f66de177e9064ee91436061c189/src/Cluster.ts but it was quite hard to debug (e.g. one thread would refuse connections in its http server from another, leading to timeouts etc), so I stopped using it, but would be nice to still see if I can get it working properly.