Mir: refactor how scopes are represented

[nikomatsakis]

Mir currently tracks scopes using these graph extents. These are a pain. They are hard to keep in sync, for one thing, as we transform the MIR. I now think we should instead store, for each statement, the innermost scope. This will allow us to recover the full set of scopes for debuginfo, as well as computing whether a statement is in an unsafe block or what.

However, for borrowck, we also need to have scopes be grouped into continuous regions of the control-flow-graph, and we need to know as we exit a scope. For this purpose, the graph extents are tolerably good, or it seems useful to have a "SCOPE-ENTRY" and "SCOPE-EXIT" sort of statement. We might just do the borrowck before we do any optimization.

In a non-lexical lifetimes world, though, introducing SCOPE-ENTRY and SCOPE-EXIT statements for all regions would be horrible and a pain. So it seems better that we keep the notion of graph extents for the purposes of borrowck, and we just avoid changing the structure between computing graph extents and running borrowck -- or else we keep the graph extents up to date.

[nikomatsakis]

cc @Doener @arielb1

[nikomatsakis]

This work is ongoing, but I think this issue doesn't add much value, closing.