Explore laziness

[cowboyd]

Graph generation can be lazy if we say that the initial entry points are fixed and then the graph cannot be changed ever again:

e.g. 10,000 users, 40,000 devices.

If we guaranteed that, then it doesn't matter if we walk the graph lazily because with same seed and same inputs, we'll get the same results.

[cowboyd]

Another possibility is that we can generate only the identity of the vertices, but not actually generate vertex data until it is requested. The rule would be that in order get the data for vertex 4, you would need to first generate the data of 1, 2, and 3 first. That way, no matter where you started asking for data in the graph, it would be guaranteed to generate the same graph based off the same seed.

Also, we would need some sort of benchmarking.