First cut/portion of stack allocation of scopes.
For now it only does that for the local storage (and only using the machine stack for it, too)

It's also somewhat buggy; I am committing to separate out the bugfix, though.

For now it's only ~3.2% speedup overall, but it shows up as 17.4% better on access-binary-trees, 16.6% better on bitops-3bit-bits-in-byte, 18.6% better on controlflow-recursive, but 14.3% slower on bitops-nseive-bits.

Boo.