Use component cache for NoFlo

[jonnor]

Takes a good bit of startup time due to being so I/O intensive.

An alternative would be to change the Processors to be long-running processes, controlled over FBP protocol. Then startup cost would only matter at worker startup, not per compute job.

[bergie]

So, two steps:

• Add noflo-cache-preheat as a postinstall step, like in https://github.com/c-base/ingress-table/blob/master/package.json#L39
• Switch to noflo-nodejs for starting the flows, run with --cache. See #56