[matthewjablack]

Summary

Lets a spliced successor contract be signed now and its funding transaction held until a later time.

• New contract_flags bit FUND_LOCKTIME_AT_CET_FLAG (0x04). A spliced contract (one with a DLC input) whose offer sets it gets a funding transaction with nLockTime = cet_locktime. ddk-dlc already turns a nonzero funding locktime into sequence 0xfffffffe on every funding input, the DLC input included, so nothing below the builder changes.
• A flag, not an argument. contract_flags already travels on the offer and is stored with the contract, so both parties, and a stored contract rebuilt later, get the same bytes.
• Old splices are unchanged. A splice without the bit, and any contract without a DLC input, keeps locktime 0. The pinned transaction digests still pass.
• Bit 1 is skipped on purpose. The stored OfferedContract reader treats a first byte of 0x02/0x03 as a compressed pubkey, so contract_flags of 0x02 or 0x03 would be misread.
• Unknown bits are now rejected. Offer creation and validation reject any bit ddk gives no meaning to, in both ddk::contract and the Manager. The builder does not reject them, so stored contracts still rebuild.

Manager

The Manager broadcast the funding transaction on sign and returned the error if it was refused. A held successor was rejected as non-final, and the contract it replaced was never marked as closing. Now:

• On sign, the accepting party stores the signed funding transaction of a held successor in dlc_transactions.fund. It is the only party with every signature.
• While that transaction is not final, the Manager doesn't broadcast. The contract stays Signed and the parent stays Confirmed.
• Finality is checked against the chain:
  • a height locktime is final once it is ≤ the tip height;
  • a time locktime is final once it is < the tip's median time past (BIP 113).
• The periodic check broadcasts the stored transaction once it is final, then pre-closes the parents. This doesn't sign again, so a signer that went offline after signing doesn't block it.
• Every broadcast error after finality still propagates.

API addition: Blockchain::get_median_time_past is a new required method. EsploraClient reads it from the tip's block info. A default implementation isn't possible: async_trait default bodies need Self: Sync, and the Manager's B::Target doesn't provide it. Any external Blockchain implementor will need to add this method.

Tests

• Unit, stateless:
  • a flagged splice has lock_time == cet_locktime, every input has sequence 0xfffffffe, both halves of the DLC input verify, and the txid matches the rebuilt transaction;
  • an unflagged splice keeps locktime 0;
  • a flagged contract without a DLC input keeps locktime 0;
  • flags 0x05 round-trip on the wire;
  • unknown flags (0x02, 0x03, 0x06, 0x08, 0x80) are rejected by create_offer and validate_offer.
• Unit, manager:
  • 0x04/0x05 round-trip through both stored OfferedContract formats;
  • 0x02/0x03 are rejected;
  • boundary cases for is_final_at.
• Regtest, stateless_execution::held_splice_broadcasts_at_the_cet_locktime: the node refuses the held funding transaction as non-final one block early, accepts the same bytes at the locktime, and the successor settles with a CET.
• Regtest, manager_execution_tests::held_splice_enum_single_oracle_test (dual-funded successor):
  • both Managers keep the successor Signed and the parent Confirmed until the locktime;
  • the acceptor's stored transaction is fully signed;
  • the acceptor broadcasts at the locktime and both sides pre-close and then close the parent;
  • the mined wtxid equals the stored transaction's;
  • EsploraClient's median time past matches bitcoind's.
• The existing splice regtest tests and the full unit suite pass.