#4227 Followups

[TheBlueMatt]

#4227 landed and I believe double-forwards HTLCs but also is a major regression in ChannelManager size. Filing this to block release, see #4227 (review)

[TheBlueMatt]

Oh, one more issue I think - the dedup code (also after #4289) doesn't consider things in the holding cells. ChannelMonitors are really only authoritative on the HTLC state if the channel is closed, not before. So because we're looking at the monitors we may double-forward still, I believe. Instead, we really need to dedup on the Channel itself. Now, maybe when we are persisting the Channel in the monitors (and are not writing the holding cell, which I think makes sense) just looking at the monitors will be right, but today it is not.