tx hash caching in transaction struct

[jchappelow]

This is a widely used approach, which Brennan has also suggested recently. Based on latest cpu profiles, we should probably do this.

HOWEVER, doing this has drawbacks, including encouraging lazy coding, and putting lots of concurrency primitives to use, which incurs its own expenses. I have seen this come to be regretted in past projects for these reasons.

First we should look carefully about when and if we should even be calling the Hash() method on the hotspots and consider other changes such as passing an existing hash and/or a new type that is simply a vehicle for a hash with it's corresponding transaction a. la the existing node/types.NamedTx.

[charithabandi]