Further points-to-based enhancements

[iamrecursion]

The passes downstream of the new points-to analysis (#264) make good use of it, but omit a number of more-complex cases as they are already big. This aims to track the "deferred" cases in both the points-to and its downstream passes (Mem2Reg, Array SROA, and Argument Promotion).

The goal of this issue is to experiment with them to see which yield measurable benefits on our corpus, and to implement those that do into the codebase as a whole.

[iamrecursion]

There are a few main knobs to explore:

• Precision enhancements for points-to that seem likely to yield downstream improvements (mostly handling globals better for now).
• Precision enhancements in Mem2reg (handling ref pointees).
• Precision enhancements in arg promotion (out-only ABI, onward-pass, 2D decomposition).
• Expanded array boundary expansion using call-graph fixpoints.

There is also a potential simplification of ripping out the steensgaard style points-to in WTI and using this one instead. This will be unlikely to change results, but will decrease maintainance burden.