Refactor as "handmade pull-stream"

[jamen]

In order to optimize this further, I could turn this process:

return pull(files, accumulator, filter, read_file)

Into:

return pull(files, reader)

Where reader would be all of them combined into a single "handmade pull-stream" (using plain functions).

Would also need to emulate functionality of pull-paramap in order to do the fast parallel reads