Add success probabilities in path finding

[thomash-acinq]

Add a new way of scoring paths that estimates the success probabilities of paths and put a cost on failed payment attempts.

For now the success probabilities are estimated in a very simple way based on capacity: I assume that the balance is a random variable uniformly distributed between 0 and capacity and multiply probabilities for all the channels on the path.
The formula for estimating the probability could be changed easily, that's not the focus of this PR. The goal of this PR is to introduce a new scoring of paths that uses this probability and can be evaluated using the new AB testing framework from #1930.

[codecov-commenter]

Codecov Report

Merging #1942 (7e0b4e2) into master (8b29edb) will decrease coverage by 0.03%.
The diff coverage is 95.74%.

@@            Coverage Diff             @@
##           master    #1942      +/-   ##
==========================================
- Coverage   87.68%   87.65%   -0.04%     
==========================================
  Files         158      158              
  Lines       12389    12406      +17     
  Branches      533      515      -18     
==========================================
+ Hits        10863    10874      +11     
- Misses       1526     1532       +6     

Impacted Files	Coverage Δ
...r-core/src/main/scala/fr/acinq/eclair/Eclair.scala	50.95% <0.00%> (ø)
...re/src/main/scala/fr/acinq/eclair/NodeParams.scala	93.92% <100.00%> (+0.14%)	⬆️
...cinq/eclair/remote/EclairInternalsSerializer.scala	97.87% <100.00%> (+0.04%)	⬆️
.../src/main/scala/fr/acinq/eclair/router/Graph.scala	98.29% <100.00%> (+0.10%)	⬆️
...cala/fr/acinq/eclair/router/RouteCalculation.scala	99.27% <100.00%> (ø)
...src/main/scala/fr/acinq/eclair/router/Router.scala	93.87% <100.00%> (ø)
...clair/blockchain/bitcoind/rpc/BatchingClient.scala	86.36% <0.00%> (-13.64%)	⬇️
...main/scala/fr/acinq/eclair/router/Validation.scala	90.76% <0.00%> (-1.54%)	⬇️
...ala/fr/acinq/eclair/balance/ChannelsListener.scala	96.55% <0.00%> (+3.44%)	⬆️

[t-bast]

When you're ready for feedback, can you update the description of this PR with details about how you plan on estimating success probabilities? I think it's worth reviewing the concept before reviewing actual code, to ensure you don't implement something that will be a concept NACK.

[thomash-acinq]

This PR is not ready yet because it depends on #1930 which is still being reviewed.

I estimate probabilities in a very simple way based on capacity: I assume that the balance is a random variable uniformly distributed between 0 and capacity and multiply probabilities for all the channels on the path.
But how I compute the probability doesn't really matter, we can change it easily. The goal of this PR is to introduce a new scoring of paths that uses this probability.

[t-bast]

Ok thanks, please update the description with that info, it's important to know!

People will be looking at our PRs but they will mostly read the description, not the code, and we don't want them to get the wrong idea on what eclair actually does inside its path-finding.

Because based on the title of the PR alone, you could have started to store historic information about failed payments, and that would have been a concept NACK for me.

[t-bast]

First pass on the general design / future extensibility.
Let me know if it's too vague, and we can discuss it to refine the proposal.

[thomash-acinq]

I've also fixed to the hopCost computation. I was only counting the last hop.