Streaming prefetch in precompute

[zachjweiner]

@inducer, taking a look here I'm realizing there's a bit of tidying that could be done (before diving into the actual implementation). Aside from the dependency on gl-284 (whose changes are unfortunately lumped into the diff here), I think I foolishly (in a rush) made some fixes to the streaming branch that really should've been applied to that branch. Also, I kinda want to rebase the changes off of master in a new branch to get rid of all the dumb merge commits.

Let me know what first steps would make this easiest. I'd be happy to lend a hand trying to expedite gl-284 (or an equivalent).