TODO

High Level

finish the grid actor scheduler
- if it's generic enough then move this package down stream to lytics/grid
add index actors - Add a required set of actors to manage each shard of an index.
1. as a proof of concept using a single actor per index
2. add a query path for the actor.
3. shard out the index over N actors,
  - add packages for shard management.

instead of using index.Index directly, is it fesable to use bleve.NewUsing("/tmp/scorch", m, scorch.Name, scorch.Name, scorchOpts)? That is the highlevel implementation of bleve? What are the implecations?
Refactor actor creation and management.
- Create a wrapper around the grid actor interface that adds some DEFAULT features to all GUI actors. Features:
  - HealthCheck messages (ping/pong): the wrapper should add a method to call actores that allow them to be pinged to see if they are active and running.
The current state machine logic in PeerState is pretty inflexible. Refactor the PeerState logic to create a DFA state machine simlar to the way lytics/dfa is used to manage the actor's own life cycle, this new dfa libary should be useful for the scheduler to manage the life cycle.
Move the relocation tests from the actor-pool package to the relocation package.
Have a look at using Jump Hash for actor locations. But it has bad distribution for low cardinality sets, see https://play.golang.org/p/JYn2IDUCLjz.