Try to handle loops in data flow more accurately in TFA

[alexmarkov]

Currently type flow analysis (TFA) breaks loops in data flow using static types (when building data flow summaries). This results in linear data flow summaries which could be applied without iterations, potentially improving speed of the analysis. However, this approximation also doesn't allow us to infer more accurate types in some cases.

We can try to evaluate what would be the outcome (both in the analysis speed and analysis results) if we would avoid breaking loops (by introducing loop or back-edge statements into data flow summary) and iterate when applying a summary until reaching fixed point.