Optimize stepping performance #67
Description
Stepping over a large number of machine instructions can easily take several seconds, which is a suboptimal user experience. We're aiming for sub-100ms response times (excluding network latency).
Stepping is currently implemented as a lazy/async action. We only advance the current machine state when requested by the user. This leaves the process idle while the user is looking at a paused state. Ideally, we should use this idle time to compute future machine states as an optimistic optimization strategy. Thus, we can serve the next state from a cache when the user requests it.