Fix broadcasting to a peer who already has tx

[randomlogin]

Previously if we broadcast to a peer who already has the tx it would hang indefinetely, as there is no subsequent getdata message.

[rustaceanrob]

This is by design. There is a possibility the first peer simply drops the inv announcement the first time we attempt to broadcast a transaction. It's the client responsibility to determine if they have already broadcast a transaction, they still want to try rebroadcasting, and to add timeouts accordingly. It would make sense to add a Builder method or direct argument on broadcast_tx to timeout.

[randomlogin]

It's the client responsibility to determine if they have already broadcast a transaction

What if we ask the same peer for the tx we've just broadcast? Of course, it doesn't mean it has propagated further, but we know that the broadcast succeeded?

And also to have a fallback with timeout?

[randomlogin]

Is there a deep purpose of broadcast_tx only erring if the node stops?
Can adding a timeout error break some guarantees? It seems not, as it could just hang.

[randomlogin]

I should not have squashed commits in this PR yet.