Remove spurious interactive-tx commit_sig retransmission

[t-bast]

We fully implement lightning/bolts#1214 to stop retransmitting commit_sig when our peer has already received it. We also correctly set next_commitment_number to let our peer know whether we have received their commit_sig or not.

Note that if our peer hasn't upgraded and always sends next_commitment_number = current_commitment_number + 1 even when they haven't received our commit_sig, this won't lead to a force-close: we will simply wait for them to send an error, which will never happen. Their channel will be stuck until they upgrade and send next_commitment_number = current_commitment_number.

This change must only be merged once:

• we've deployed our node with the changes from Reestablish partially signed splice with current remote_commitment_number #2965
• we've made a Phoenix release that includes Correctly set next_commitment_number during splice reconnect lightning-kmp#736 so that users can upgrade to avoid being stuck

[pm47]

That's subtle, hard to be sure all combinations are correctly handled, but LGTM.

[remyers]

How do we test the behavior "...if our peer hasn't upgraded and always sends next_commitment_number = current_commitment_number + 1 even when they haven't received our commit_sig, this won't lead to a force-close: we will simply wait for them to send an error, which will never happen." ? If both Bob and Alice send commit_sig and Bob does not receive it but sends bobCommitIndex + 1 in their channel_reestablish (legacy behavior), then after a reconnect would we expect Alice to wait for Bob to send tx_signatures and Bob to wait for Alice to send commit_sig again? Perhaps a test that shows their hung state and how a reconnect with the correct commit index from bob get's them to be unstuck.

[t-bast]

Perhaps a test that shows their hung state and how a reconnect with the correct commit index from bob get's them to be unstuck.

I have rebased to fix conflicts and added de331c8 where I showcase the behavior with non-upgraded nodes.

[t-bast]

I'll ask feedback during tonight's spec meeting before merging this PR. If other implementations are fine with this, we'll merge this PR and make an eclair release tomorrow.

[t-bast]

We've discussed this during yesterday's spec meeting: we have an ACK to release, so I've done a last rebase and we can now merge!

[remyers]

LGTM!