Move router handlers to separate files

[pm47]

This build on #1347.

The goal is to simplify the router by moving handlers to dedicated stateless functions in separated files, for syncing, announcement validation, route calculation, etc. That's quite a big change, but it's very repetitive.

Also, all gossip messages are now acknowledged by the router with a GossipDecision reply.

[codecov-io]

Codecov Report

Merging #1352 into master will increase coverage by 0.23%.
The diff coverage is 94.37%.

@@            Coverage Diff             @@
##           master    #1352      +/-   ##
==========================================
+ Coverage   86.40%   86.64%   +0.23%     
==========================================
  Files         119      123       +4     
  Lines        9306     9364      +58     
  Branches      390      384       -6     
==========================================
+ Hits         8041     8113      +72     
+ Misses       1265     1251      -14     

Impacted Files	Coverage Δ
...air-core/src/main/scala/fr/acinq/eclair/Logs.scala	89.79% <ø> (+2.04%)	⬆️
...re/src/main/scala/fr/acinq/eclair/NodeParams.scala	88.33% <ø> (ø)
.../src/main/scala/fr/acinq/eclair/db/NetworkDb.scala	100.00% <ø> (ø)
...src/main/scala/fr/acinq/eclair/db/PaymentsDb.scala	80.00% <ø> (ø)
...la/fr/acinq/eclair/db/sqlite/SqliteNetworkDb.scala	97.29% <ø> (ø)
.../scala/fr/acinq/eclair/payment/PaymentEvents.scala	100.00% <ø> (ø)
.../scala/fr/acinq/eclair/payment/PaymentPacket.scala	90.90% <ø> (ø)
...scala/fr/acinq/eclair/payment/send/Autoprobe.scala	0.00% <ø> (ø)
...clair/payment/send/MultiPartPaymentLifecycle.scala	98.27% <ø> (ø)
...r/acinq/eclair/payment/send/PaymentInitiator.scala	95.89% <ø> (ø)
... and 23 more

[t-bast]

This is a great clean-up!

I think we can go further, but if we do can you make sure you don't rebase so that we can keep the history of what we've already reviewed (these refactorings take time to properly review, so better to do them only once!).

Or we can choose to do follow-up PRs for some of my comments.

[t-bast]

Maybe it's worth defining a trait next to GossipDecision to flag which cases should lead to a connection close?
This way we'll make sure anytime we update the list of potential gossip decisions in router, we correctly flag the new ones?

[pm47]

But we're not doing the exact same action. Another way would be to just replace the case _ => catch-all by more precise matchers. Compiler would warn us if we're missing a new type.

[t-bast]

But couldn't this catch-all be expressed by a specific trait defined in the router?

[pm47]

Only if the trait meant "should not lead to a connection close".

[pm47]

I've moved all static methods to respective handlers file, and also took the opportunity to move router-related types to the Router object.

As this touches a lot of files and may generate conflicts, I suggest we merge this rather sooner than later.

[t-bast]

LGTM, thanks for this!

[pm47]

Rebased, waiting for review from @sstone.

[sstone]

At first I did not like passing an akka context to most helper methods, which means they are not "pure", but this makes Router much more readable and it is still possible to tests all side effects cleanly.
A "pure" way of doing this would be to have methods that return the new state and all side effects but we would need to model them all (ack message, reply to sender, publish event, send message to self, send message to another actor...) and it's probably overkill for now but I'll open an issue to keep that in mind.

[t-bast]

I completely agree with @sstone, that was exactly my reaction too :)
It's a good first step towards that direction, and doing everything in one step would have been too much.