key-wallet: islock conflict-sweep and abandon removals emit no events — mirrors keep dead rows, released coins stuck spent

[bfoss765]

Found by the 2026-08-19 merged-PR audit (mechanics introduced/exposed by #961/#962). Three related defects, all the same engine-vs-store divergence class:

1. islock conflict-sweep emits no removal event (HIGH): mark_instant_send_utxos discards the WalletConflictSweep result, so when an islock conflicts a wallet transaction the engine sweeps it internally but TransactionsSwept never reaches the FFI event stream. Host persistence mirrors (Kotlin/Swift) keep the dead tx rows, and coins the sweep released stay marked spent forever. This is the normal tx-then-islock ordering, not an edge case.
2. Released outpoints never restored to core utxos (MED-HIGH): the consumer is told the coins are spendable again while coin selection cannot actually spend them; the restore path is unreachable once the funding tx is chainlocked. (Raised by QE on fix(key-wallet): stop a never-broadcast transaction from crediting money that does not exist #961 thread 2; merged unresolved.)
3. Abandon path performs removals with no event (MED): same divergence class.

A complete fix exists on branch fix/sweep-event-emission (former PR #971, closed to keep the open-PR queue focused on migration-critical work): events wired through process_instant_send_lock, plus re-credit gated on a chainlock-horizon (spend_proof_horizon) so a re-credited coin can never race an in-flight spend proof. Reopen the PR as-is when there is review bandwidth.

Note: this divergence class overlaps the layer-1 reconcile planned for the restore-persistence bug — it may be simplest for that reconcile to heal it store-side, with this branch as the engine-side complement.