Dominator tree

[stoklund]

Cretonne's in-memory IR needs a dominator tree analysis for verification, and likely for register allocation.

We expect most input programs to have structured control flow, so the Cooper / Harvey / Kennedy algorithm will work well.

• Should the dominator tree be a static analysis, or should it be dynamically updated when the CFG (Control flow graph #1) changes?
• How does a dominator tree interact with EBBs? The immediate dominator of every inner basic block in an EBB is obvious (it's the previous basic block), so it doesn't need to be stored. Inly the immediate dominator of an EBB is interesting.

A rough draft: https://github.com/mrrrgn/cretonne/commit/f6dd0270d0f8f09e82c7a6be74548cd6ec69f15f

Closing this for now. https://github.com/stoklund/cretonne/blob/master/src/libcretonne/dominator_tree.rs