WIP: Draft: Splicing

[JssDWt]

Under the following conditions the lsp will splice in a channel, rather than opening a new channel:

• The lsp node supports splicing (cln with --experimental-splicing enabled)
• The peer supports splicing
• The lsp has a zero-fee-htlc-tx channel with the peer

If above conditions are true, the LSP selects a channel to splice in based on some rules, to try and make sure always the same channel is used to splice-in over. Basically using the oldest zero-fee-htlc-tx channel, or the 'lowest' tx hash if all channels are unconfirmed.

TODO:

• Htlcs are not yet able to be forwarded over the newly spliced channel. Cannot forward htlc over spliced channel? ElementsProject/lightning#7003
• Miner fees need a rework. The wanted feerate is passed to cln, but the resulting tx has a lower fee rate than requested. It's not possible to parse the psbt with the btcd psbt library in order to deduce the fee ourselves, because btcd considers the psbt invalid: Unable to decode psbt produced by splice_init, splice_update and splice_signed ElementsProject/lightning#7004
• The output index of the 'new' funding transaction is unknown. That is needed in order to be able to forward to the correct destination, and for internal storage. Splicing: how to get the outnum of the new funding transaction? ElementsProject/lightning#7005
• 0-conf splices are not possible with the current version of cln out-of-the-box. Allow 0-conf splices ElementsProject/lightning#7002 (needs spec work?)
• Remove the chain confirmations from the test.

Caveats:

• If the splice funding fails, the channel may forever be in stfu/quiescence mode. Making it impossible to use again. I'm not sure yet whether there is a way out of that.
• If htlcs are in-flight on the channel when trying to splice-in, the splice-in would fail and thus the htlcs that would pay for the splice-in would fail too.

Notes:

• Zero conf splices can be chained. So there can be a chain of unconfirmed splices in the mempool, just like with regular channel opens. We have to be careful that the lightning implementation does not try to RBF the splice transaction, because that would make the channel invalid. Probably they won't do that after splice_locked in sent, which should be immediate for 0-conf splices.

[ddustin]

The enable splicing config option was moved to runtime -- it's set the the conf file or runtime parameter now