Send transactions to all peers instead of a sub-set

[ricardolyn]

Summary

The introduction of EIP-2464 (eth65 version) reduces the bandwidth used for transaction propagation from linear complexity in the number of peers to logarithmic one. While this approach removes the naive transfer of full transactions, it does create transaction pool consistency issues where some nodes are unable to promote transactions properly to be included in a block and thus creating empty blocks.

Moreover, it happens more specifically when there is a combination of the following:

• high throughput of transactions from the same EOA
• lower value than throughput on Transaction Pool Options (like --txpool.accountqueue)
• a low number of nodes in the network (the lower, the higher probability that the node is unable to promote remote transactions).

The following image illustrates the issue:

Therefore, this PR addresses the issue by sending the transactions to all peers, while in the meantime we work on a logarithmic propagation solution for IBFT/QBFT.

Changes

• Broadcast the full transaction data to all peers
• Add unit test to validate the logic

[baptiste-b-pegasys]

I know it is inspired from code that exists priorly, but I just want to raise some concerns here about closing and emptying channels, checking all errors, and trying to be synchronous to avoid some flakiness.

[ricardolyn]

@baptiste-b-pegasys changes done. please review again

[jbhurat]

LGTM

[achraf17]

LGTM