Test for same-preimage different-direction HTLCs on one channel on-chain

[TheBlueMatt]

Looking at it, check_spend_remote_transaction may be wrong in the non-revoked case when we have two HTLCs with the same preimage that went across a node in opposite directions. Specifically, it looks like if we have a preimage, we sign an input spending the output, even if that output is an outbound HTLC and our signature will be bogus (potentially making the whole combined-claim transaction invalid).