Add progressive file reading.

[andrewgross]

We currently read the whole file in to memory at once, and build our representation in memory. This is not great with large CSV files. We should instead read in chunks of the file at a time and progressively grab more as needed. I think this can be done with double generators.