[ajaysehwal]

Fixes #1140.

get_mempool_transactions_and_timestamp_at_height decided whether a mempool transaction had already been emitted only by comparing its entry time to the saved cursor. Bitcoin Core's entry times have one-second resolution, so a new transaction in the same second as an already emitted one was skipped, and stayed skipped while it was unconfirmed. The wallet then missed the unconfirmed payment until it confirmed.

We now track emitted txids instead:

• An entry is skipped only if its txid was already emitted and its ancestors are known to be processed (the existing entry.height <= best_processed_height check is unchanged).
• Txids that left the mempool are forgotten, so a transaction that re-enters is emitted again. The set is bounded by the mempool size.
• Txids are added only after a successful poll, so a failed poll doesn't mark transactions as emitted that the receiver never got.

The timestamp cursor is still used for evicted-transaction timestamps, and the body cache is unchanged.

Tests: the selection logic is a separate function with unit tests for the same-second case, no repeats, lagging ancestors, eviction and re-entry, and a failed poll not being remembered. The same-second, re-entry and failed-poll tests fail with the old rule. The lib tests, cargo +nightly fmt --check and cargo clippy --lib -- -D clippy::unwrap_used pass.

AI disclosure: prepared with the assistance of Claude Code.

[ldk-reviews-bot]

I've assigned @tnull as a reviewer!
I'll wait for their review and will help manage the review process.
Once they submit their review, I'll check if a second reviewer would be helpful.