Make ClaimHtlcTx replaceable

[t-bast]

We add claim-htlc transactions to the set of transactions that we will RBF if they don't confirm quickly.

This is important because the feerate is evaluated when we see the remote commitment in the mempool, so it may be outdated by the time we can actually publish the claim-htlc transaction (either because of the absolute timelock or because the remote commitment took a long time to confirm).

It's very simple to set the fees for these transactions: since we can unilaterally sign, we simply decrease the output amount and re-sign. Note that we do it for both standard commitments and anchor output commitments. For standard commitments, these transactions are marked as non-RBFable, so RBF attempts will simply fail if the previous transaction is in the mempool. But it's still worth attempting it, as it may have been dropped from the mempool in high-fee environments.

The ReplaceableTxPublisher deserves some refactoring: integrating these various types of transactions shows where the model works and where it doesn't. But I've got a few other changes I'd like to make which impact the architecture, so I'd like to keep it like this for now and refactor after a few more PRs.

[pm47]

I like how easy this change is! Shows that our model is good.

The ReplaceableTxPublisher deserves some refactoring

A few minor remarks:

• How about renaming RawTxPublisher to FinalTxPublisher?

• eclair/eclair-core/src/main/scala/fr/acinq/eclair/channel/publish/TxPublisher.scala

  Line 187 in 0b807d2

  if (alreadyPublished) {

  Wonder if this precheck should be done when starting the RawTxPublisher actor, which would then stop right away (same for ReplaceableTxPublisher).

• eclair/eclair-core/src/main/scala/fr/acinq/eclair/channel/publish/TxPublisher.scala

  Line 207 in 0b807d2

  log.info("not publishing replaceable {} spending {}:{} with feerate={}, publishing is already in progress", cmd.desc, cmd.input.txid, cmd.input.index, targetFeerate)

  This log should make it clear that the existing tx has a higher feerate and that's why we don't replace it (e.g. append " with higher feerate={}").

[t-bast]

How about renaming RawTxPublisher to FinalTxPublisher?

Good idea, it's a better distinction with ReplaceableTxPublisher.
Even though I have in the back of my mind ideas for a future where such "final" txs can be CPFP-ed through their change output (e.g. for funding transactions), but whether it's Raw or Final doesn't matter for that.

I'll do this in a follow-up PR.

Wonder if this precheck should be done when starting the RawTxPublisher actor, which would then stop right away

We can't do that for the ReplaceableTxPublisher, because the decision is based on the map kept by the TxPublisher.
It's true that we could do it for the RawTxPublisher, but I don't like creating this asymmetry, I think it's better to do it at the same place for both types of publishers (so it has to be in the TxPublisher).

This log should make it clear that the existing tx has a higher feerate and that's why we don't replace it

In the next PR (#2113, not yet completely ready for review) I change this to use a deadline instead of an explicit feerate, so I'll fix it there.