cryptonote_protocol: prevent duplicate txs in fluff queue

[0xFFFC0000]

Fix duplicate transaction #9335

Thanks @Boog900 for suggesting how it can be done easier.

There is another (draft) implementation which passes tests based on std::set here.

[vtnerd]

As a clarification - is this a race condition on the txpool? If the tx is already in the pool, it shouldn't re-relay it. But this should be possible if two connections fluff nearly at the same time.

[Boog900]

@vtnerd This happens when the tx pool size limit is set too low and you start dropping transactions too soon after receiving them.

When I tested this I set the tx pool limit really low, then sent 2 separate fluff messages of the same tx one after the other (not at the same time) with an extra (not duplicate) tx in the first message, then on another connection monerod sent a duplicate tx in the fluff, which would cause peers to disconnect.

I think this is the cause of monerod struggling to keep connections with a low max txpool size.

[vtnerd]

I keep forgetting about that limit. This fix (once corrected) should keep the connections alive from what I've seen (just re-reviewed the tx relay code).

[selsta]

@0xFFFC0000 please also open against release branch