Develop a cost model

[lutter]

The spec for the read interface does not define the cost model for queries. That model needs to be developed based on measurements of representative queries, and taking the non-query overhead that an index node incurs into account.

[Jannis]

We're doing it but not tracking it via a graph-node issue, since it's outside graph-node anyway.