network: Handle failed to broadcast transactions

[jsgranados]

Summary

When a txn accepted by a node (both REST and WS) first we Remember and then attempt to broadcast. Broadcast itself can fail for various reason including a pure network performance errors.New("broadcast queue full") reason.

In this case the txn could stuck on a node (npn, part) until it expires.

Solution options to consider:

1. Modify txsync to pull from nodes and re-broadcast to nodes excluding the node pulled from (X)
2. Handle errBcastQFull from Broadcast/Relay in node/txHandler and try again after some delay later (S)
3. Make txpool to maintain non-flushing txns, and re-broadcast them every 2-3 (N) rounds: after 2+ rounds local txns are expected to be included into some proposal. Note, in p2p world N could be > 3+ (M).

Acceptance

1. Unit test

Urgency

This is a well known issue. Some partners (i.e. Circle) encounter this.
It would also improve overall txn sending expedience.

[jsgranados]

Approach:

• Make txpool to maintain non-flushing txns, and re-broadcast them every 2-3 (N) rounds: after 2+ rounds local txns are expected to be included into some proposal. Note, in p2p world N could be > 3+ (M)
• Disambiguate TransactionPool.Remember: transaction already in ledger error as it is misleading due to the transaction not being in the ledger, but in the pending block evaluator