Stub out methods at parse time

[beckjake]

Stub out many adapter methods at parse time. Unfortunately this doesn't appear to reduce the integration test runtime, but I've verified it does prevent calls to the various adapter methods from running (with a real project) and added unit tests about it. It nicely solves the problem of needing to parse files before being able to populate the cache by stubbing out everything that might want the cache at parse-time.

already_exists returns False for consistency with get_relation.

I'm not sure this is the most perfect way of doing it, but using it does feel reasonably pleasant and it seems more sustainable than overriding it manually inside the parser or something. I stole the available.parse = available_parse idea from mock's patch.object.

I think the biggest downside is that you have to decide whether or not to stub it out in parsing when you mark them available, not at actual implementation time, because the wrapper only delegates the top-level call, and subsequent accesses to self hit the wrapped adapter only. I think that's pretty much fine, but I did want to call it out.

[drewbanin]

This LGTM! There are some pretty strong assumptions that have motivated this change.... let's be sure to test the heck out of this on a variety of representative dbt projects.