tx-submission decision logic for the inbound side

[coot]

To smooth traffic and save bandwidth, we can add a decision logic to inbound
tx-submission protocol which would decide from whom to download an available
transaction and trace which transactions are being downloaded (so we don't
start to download something that's already being downloaded).

todo list

• validate tx size
• validate that tx size is smaller than maximum tx size
• efficient adding txs to mempool (do not try add the same tx twice)
  Is this even needed? Mempool could check txids before doing any expensive validation.

[dcoutts]

This would be a design change with the following benefits:

1. substantially reduced bandwidth for tx broadcast
2. smoothing out refilling of the mempool to reduce bursty CPU and network use

This will be a useful optimisation as we get towards higher and higher chain throughput / TPS. For the moment (at current low TPS rates) it is not a significant issue. The effects are only visible in benchmarks with very high TPS.