[kwsantiago]

After a mutual close, lightningd watches each pending funding candidate (a splice, or an earlier dual-funding attempt) and force closes if one of them confirms instead of the close. The callback passed the candidate's transaction to the failure path as if it were a commitment that had already closed the channel, so drop_to_chain() logged "Not dropping our unilateral close onchain" and published nothing. The channel stayed in CLOSINGD_COMPLETE with its funds in an output nothing was watching, until a restart.

The callback now adopts the candidate the same way funding_spent() does (crediting the splice amount and saving the channel), then fails the channel without a transaction, so our commitment on the new funding output is published. It does nothing if the candidate was already adopted, for example when a reorg mines it again.

channel_apply_scid() no longer changes an existing scid once the channel is closing. After a restart, the funding watch on the adopted output computed a new scid, and changing it on a channel gossip already considers dying logged a broken gossip state transition.

Tests:

• test_splice_confirms_after_mutual_close (v1 and dual-funded): a splice confirms in place of the mutual close; checks the funding, the balance including the splice-in, a restart within the rescan window, a restart past it that rebroadcasts our commitment, and the onchain resolution. It fails without this change.
• test_rbf_candidate_confirms_after_mutual_close: the first dual-funding attempt confirms in place of the bump the close spends. It fails without this change.

Changelog-None