Investigate making virtual registers trace local

[pgoodman]

This would require modifying the fall-through of a function call to always be a kRequestBlockInFuture, and would also mean that if any fragment's stack is invalid, then all fragment stacks are invalid. The benefits would be that it makes VRs slightly less restricted, and would potentially mean less added instructions (and code!) because I wouldn't need to partition fragments anymore. I think this is definitely worth investigating.

[pgoodman]

Note: The new "headless" compensation blocks will have to be actual fragment boundaries. Fall-throughs of system/interrupt/function calls will have to be implicitly considered headless.