Refactor bookstore to load data in very few queries (#40)

## Product change and motivation

The `bookstore` sample dataset now loads its data using very few queries
(31, down from about 310).

This allows HTTP-based clients to load it very quickly without getting
bottlenecked hard by network latency.

## Implementation

By making the `match` stages much chunkier we can drastically cut down
on the total number of `insert` stages required to insert all data. For
some types, we go from N queries to 1 for insertion of all instances of
said type.

Author: alexjpwalker