WIP: Eliminate artificial transaction delays
This is an experimental change proposal, being put up for discussion. I am not claiming it has been fully considered - that is why I'm putting it up for debate. It implements (as far as I can tell) the request that was forwarded from Emil Oldenburg in #5.
The changes here currently pass extended regtests, but is just a first phase and may be altered radically or abandoned entirely depending on the outcomes of discussion.
Cynics would ask how much of the original code in critical sections can just be deleted without any test failing.
- think about whether to scrap the delays, or configurable down to zero (in which case a test and further decoupling from INTERVAL_BROADCAST_MAX_PER_MB is needed)
- have a system-level regtest benchmark for measuring the latency?