Proper On Stack Replacement

[mgaudet]

Currently we manage JIT to Interperter transitions by restoring all interpreter state. We do this because JIT frames exist on the C-stack, along side interpreter frames, which is a problem when Ruby's exception handling story is implemented partially through setjmp and longjmp

To improve performance, it would be nice if the JIT only had to restore interpreter state when it absolutely had to. In order to do this, we need to implement proper on-stack replacement and connect it together with the exception handling system.

Before longjmping past a JIT frame, it needs to be given an opportunity to restore state it may have privatized, or we need to get the JIT frames off of the regular C call stack, and manage them ourselves.