Functionalization pass v1

[bdhirsh]

First pass of the functionalization pass for functorch. Most of the description can be found in the accompanying core-side PR here.

This PR probably needs some cleanup first, but I wanted to put up the functorch-side changes early for completeness.

Stuff that I still need to do:

• I've only done basic testing of functionalize() so far
• I need to implement more view ops - only view() is currently implemented

The part worth talking about the most is probably the wrapping/unwrapping logic here, which is kind of similar to the wrapping/unwrapping logic for grad() (and can probably be shared). Main points:

• functorch needs to do the wrapping/unwrapping, and not core, because core doesn't know what "level" the wrapper should have. That means that the wrapping and unwrapping should happen in the boxed fallback, DynamicLayerBackFallback.
• It needs to be careful to wrap the outputs of factory functions, but not to wrap any outputs when we're currently printing.

[zou3519]

I'm cleaning up old PRs in preparation for the functorch->pytorch merge; we've already landed functionalization in some form so closing this.