Revamp borrowck treatment of closures

[nikomatsakis]

As described in #2202 and this blog post, the borrow checker's treatment of closures needs to be changed. When a new closure is created, we should analyze the upvars that it accesses and create suitable borrows of those upvars that persist for the closure's lifetime. This is interrelated with #3696.

[nikomatsakis]

Part of #2202

[metajack]

triage bump. @nikomatsakis should this and the related bug be nominated? How does this interact with @fn being gone, etc?

[lilyball]

If I understand this correctly, fixing this will significantly impact anyone who's using a closure in a destructor for RAII purposes (e.g. #11905). Nominating for backwards compat.

[pnkfelix]

Accepted for 1.0, backcompat-lang.

[nikomatsakis]

I've got a patch for this nearly ready to land. It definitely impacts some existing code. The patch is code complete but I've been working on writing up docs on the affected code patterns and how they must be updated.