ProbabilisticScorer learning-from-probes interface

[TheBlueMatt]

While you can do it already, passing information about a successful probe into ProbabilisticScorer is somewhat confusing today - you have to do it by calling the payment-failed method with a bogus scid to "blame". We should just wrap that call in a util function and expose it officially.

cc @jkczyz

[jkczyz]

We may want to expand this issue to implement a background prober, which would actively probe paths in NetworkGraph with various amounts to learn the liquidity balances. It would need an EventHandler that could distinguish its probes from actual payments. In the former case, it would need to call the scorer with a fake failed scid as noted. Otherwise, delegate to the actual event handler.

[TheBlueMatt]

Another behavior I'm observing with the new scorer (at least with the settings on latest git) - it sometimes gets really excited when it finds a path that works, and tends to strongly prefer things it is confident will work over things its unsure about. that's probably a great outcome during normal payments, cause it seems to often be correct, but when we build a first-class prober, we should definitely make sure the prober is forced to probe each available channel we have outbound capacity on, instead of getting stuck using one, should be easy to do, though, just filter the ChannelDetailss we pass through and walk through them as we probe.

[TheBlueMatt]

Oops, 1567 didn't actually end up adding any prober logic in the scorer, which was the point of this issue.

[TheBlueMatt]

This has been done for a while.