Investigate ways to make generated code more stable

[aborg-dev]

There have been a few cases when ZKASM codegen behaved in a non-stable way and generated a different code (sometimes more and sometimes less efficient), e.g. see #187 (comment)

This causes unnecessary diffs in PRs and makes tracking effects on performance harder - we need to find a way to make the produced code more stable.

It could be useful first to nail down when this instability happens and then see how different compilation flags influence it.