SSA pass

[kripken]

This adds a pass to move to SSA form, i.e., each local has a single assign. (See comment in pass source about phis, we don't add a phi node here.)

This has a worst-case runtime of max-loop-depth * max(num-locals, num-uses) * max-defs-for-local. I don't know how that compares to typical ways of doing SSA reconstruction - curious if anyone knows offhand? Anyhow, it seems like a reasonable worst-case.

Goals of this PR:

• Improve our ability to optimize partially-optimized code from other compilers, like the wasm backend, or input not in SSA form, like AssemblyScript. We'll need more tools here before it makes sense to measure in depth, I plan to get to that next.
• A possible step towards a full SSA+phis IR in binaryen, which as discussed in the past, we may want in the future.

This PR also includes a few tiny bug fixes that this uncovered, and improvements that made doing this easier. That includes an InstrumentLocals pass, like InstrumentMemory but for locals, which was very useful to debug this.