Add initial support for async payment trampoline relay

[remyers]

This is the first step in adding "async payments" as described in issue #2424 .

[remyers]

Some Questions for my next set of changes to this PR:

• What information should go in the AsyncPaymentFeatures onion tlv?

I currently use invoice feature bits because an empty tlv seemed wrong. The sender will look for the AsyncPaymentPrototype feature bit in the invoice, but the trampoline relay does not have access to the invoice (for privacy). That's why the sender must add a tlv specifically to the onion of the trampoline relay.

• When I change the AsyncPaymentTrigger message to be a non-spec LightningMessage that is routed to the NodeRelayer, should it be keyed off of the nodeRelayPacket.outerPayload.paymentSecret for the node relayer ?

• Would it be better to publish an event when an AsyncPaymentTrigger non-spec LightningMessage is received, or route it to the set of running NodeRelayer actors ?

[t-bast]

Nice, you found your way around the right parts of the codebase, most of this looks correct!

When I change the AsyncPaymentTrigger message to be a non-spec LightningMessage that is routed to the NodeRelayer, should it be keyed off of the nodeRelayPacket.outerPayload.paymentSecret for the node relayer ?

I don't think we need to define a LightningMessage for the AsyncPaymentTrigger.
This won't be something we receive from the recipient peer when it comes back online.
It will just be an internal component that will send this internal message to the event stream.

Let me know what you think, I think the flow should be the following:

• we should add a singleton actor named something like AsyncPaymentWatcher
• when the NodeRelay component goes to the waitForTrigger behavior, it should send a NotifyWhenPeerOnline(replyTo, remoteNodeId, paymentHash) to the AsyncPaymentWatcher
• the job of the AsyncPaymentWatcher will be to check when the peer comes back online, and when the peer is back online, to send AsyncPaymentTrigger/RelayAsyncPayment to the corresponding replyTo actor

What I don't know yet is if this should be a new actor or if we should enrich an existing actor with that new responsibility.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.86%. Comparing base (ba2b928) to head (f670676).
Report is 318 commits behind head on master.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2435      +/-   ##
==========================================
+ Coverage   84.71%   84.86%   +0.15%     
==========================================
  Files         199      198       -1     
  Lines       15748    15796      +48     
  Branches      656      680      +24     
==========================================
+ Hits        13341    13406      +65     
+ Misses       2407     2390      -17     

Files with missing lines	Coverage Δ
...core/src/main/scala/fr/acinq/eclair/Features.scala	99.27% <100.00%> (+0.02%)	⬆️
...re/src/main/scala/fr/acinq/eclair/NodeParams.scala	93.26% <100.00%> (+0.12%)	⬆️
.../scala/fr/acinq/eclair/payment/PaymentEvents.scala	97.61% <ø> (ø)
...cala/fr/acinq/eclair/payment/relay/NodeRelay.scala	96.55% <100.00%> (+0.61%)	⬆️
.../scala/fr/acinq/eclair/payment/relay/Relayer.scala	89.13% <ø> (ø)
...a/fr/acinq/eclair/wire/protocol/PaymentOnion.scala	99.25% <100.00%> (+0.02%)	⬆️

... and 15 files with indirect coverage changes

[remyers]

RE: @t-bast 's point earlier about the possibility of a channel closure if the receiver does not trigger the payment and the sender is not back online within cancel-safety-before-timeout-blocks of when it times out.

An async payment sender should probably set the payment expiry well beyond their LSP's expected hold-timeout-blocks (default: 1008 blocks, ~7 days). If the default async payment expiry was something like twice the default hold-timeout-blocks, then the sender would need to be online once within a 6 day period before the payment expires to prevent a channel closure. They would also have a 7 day period before the LSP's timeout to manually cancel the payment.

[t-bast]

LGTM, a last couple of nits and we should be good to go 🚀

[t-bast]

LGTM! Don't forget to use github's "squash and merge" option, this is good to go 👍