Reduce witness gen code bloat by supporting runtime witness gen loops

[clvv]

Problem

Currently, repeated circuit components incur large witness generation code. This is because there is a separate line of rust statement invoking the witness generator for each circuit component each time the component is used. This results in long compilation times for the generated witness gen code.

Solution?

We need to explore ways to reduce the generated code size. Effectively, iterative calls to the same witness gen needs to be generated as runtime loops, instead of "unrolled" loops as they currently are. Moreover, we should support both sequential and parallel runtime loops.

The exact solution is TBD.