Canonical block transaction-ordering

[surg0r]

Background
It would be useful to prepare for higher transaction loads on the network with eventual implementation of set reconciliation and invertible bloom lookup tables (IBLT). The general idea is that all nodes on the network see the same transactions and therefore when block transaction counts rise there is little point in wasting bandwidth transmitting the full block and all the transactions together. This can be massively optimised - see graphene and this.

Suggested interim upgrade
To implement this protocol upgrade we require a minor change to transaction ordering in the block. Currently transactions are ordered by transaction fee but I propose we update this to instead order by a unique key such as the transaction txhash identifier. Such a change and implementation of a version of Graphene would be not be a consensus-rule change and therefore would not require a hard fork (although some handshake protocol exchange parameters will need to be updated)..