Date mempool absences with our own clock

[tnull]

The bitcoind chain source dated a transaction's absence from the
mempool with the newest mempool entry time it had seen since startup.
That watermark restarts at zero while the transaction's `last_seen`
persists with the wallet, and BDK keeps an evicted transaction
canonical until its eviction catches up with `last_seen`, so a
transaction that vanished while we were down kept its input marked
spent and its change counted as ours.

Date the observations we hand BDK with our local clock instead, as the
Esplora and Electrum sources and upstream `bdk_bitcoind_rpc` do, and
keep the entry-time watermark for emission deduplication only.

This only bites after a restart, and only while the mempool holds
nothing newer than the transaction: otherwise the first poll re-emits
the whole mempool and advances the watermark before that same poll
reports any eviction. In practice that means an empty mempool on
signet or regtest, or a similarly idle backend, and the next
transaction to arrive resolves it anyway.

[ldk-reviews-bot]

I've assigned @tankyleo 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.

[tankyleo]

I raise this point after a chat with codex: this patch can cause the BDK last_seen to be equal to evcited_at in case two polls happen within the same second, and the first poll observes an eviction, while the second poll observes the transaction's return. In this case BDK keeps the transaction evicted, and subsequent polls never bump last_seen again because the entry is now at or below the monotically increasing latest_mempool_timestamp.