sweep: group pinnable and non-pinnable inputs separately

[morehouse]

The sweeper currently has no concept of pinnable and unpinnable inputs, so HTLC-Timeout (pinnable) and HTLC-Success (not pinnable) inputs can be aggregated into the same transaction if they have the same deadline. As a result, the HTLC-Success inputs effectively become pinnable too.

Another example of this is expired anchors (pinnable) being aggregated with HTLC-Success (not pinnable), though this could also be solved by removing expired anchor sweeping.

I think @yyforyongyu's proposed "sweeper groups" would be all we need to properly implement this.

[TheBlueMatt]

We did this in LDK in the latest release (I assume that's where @morehouse saw this). One subtle issue, though, is an HTLC can become pinnable once it (gets close to) expiring as suddenly both parties can claim it. Thus it's useful to track the pinnable state rather than making it fixed.

[morehouse]

@TheBlueMatt Will transactions actually relay with an nLockTime in the future? If not, then this isn't a problem until the HTLC actually expires, in which case we're partially screwed already. Of course, splitting expired inputs out at that point helps control the damage.

[TheBlueMatt]

They will not, no, indeed, for non-revoked transactions its really only an issue after expiry. If you do this for expired counterparty broadcasts, though, it needs consideration.