Handle splice with local/remote index mismatch

[t-bast]

We assumed that once quiescent, the local and remote commitment index would be the same, but that's not true. Those two indices may diverge because of concurrent updates. We need to keep using the right index for each commitment during and after a splice, otherwise it leads to "invalid commit sig" errors.

[codecov-commenter]

Codecov Report

Merging #2757 (8bb9fa6) into master (0e4985c) will increase coverage by 0.06%.
The diff coverage is 100.00%.

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

@@            Coverage Diff             @@
##           master    #2757      +/-   ##
==========================================
+ Coverage   85.81%   85.87%   +0.06%     
==========================================
  Files         216      216              
  Lines       18070    18070              
  Branches      732      751      +19     
==========================================
+ Hits        15506    15517      +11     
+ Misses       2564     2553      -11     

Files	Coverage Δ
...c/main/scala/fr/acinq/eclair/channel/Helpers.scala	94.33% <100.00%> (ø)
...inq/eclair/channel/fund/InteractiveTxBuilder.scala	91.47% <100.00%> (ø)

... and 4 files with indirect coverage changes

[remyers]

The fix and new test look good. I also confirmed the test fails without the fix.

[remyers]

In the process of testing I discovered a small bug in the crossSign helper which I fixed in ab4e7ec. Would it make sense to wrap that fix into this PR, or create a new one?

[remyers]

Do you think testing with mismatched commit indexes would be useful for all of the quiescence splice tests?

After some false starts, I modified setupHtlcs to test with mismatched commit indexes in 927e114.

Both changes are in branch https://github.com/remyers/eclair/tree/splice-remote-commit-index

[t-bast]

In the process of testing I discovered a small bug in the crossSign helper which I fixed in ab4e7ec.

Very good catch, thanks! There was another bug as well (related to having mutiple commit_sig), which I patched in ad5e84d

Do you think testing with mismatched commit indexes would be useful for all of the quiescence splice tests?

No, that would be overkill. That's a specific edge case we want to test once, it doesn't need to be the code path we use in all tests (especially since it's way less readable).

[remyers]

LGTM!