Lower CGGI operations on high level integers to LWE/PBS

The `--arith-to-cggi` pass lowers arithmetic operations on larger bit width integers to CGGI operations exposed by shortint / integer APIs. This is an alternative non-Yosys pass that can enable much faster compilation and execution than circuit booleanization.

For example, this would include `cggi.add`, `cggi.sub`, `cggi.mul` and bitwise operations like `cggi.xor` `cggi.and`.

This could belong in a pass like `--cggi-decompose-operations` to decompose higher level CGGI operations into scalar LWE ops and PBS ops that hardware accelerators can then schedule and execute.

In particular, starting with a `cggi.xor` operation assuming 2 bit messages and 2 bit carry space (so that LHS and RHS can be concatenated with scalar ops and then a 4-bit PBS can compute the XOR truth table) would be a great first step.

This would help enable a comparison between hardware accelerators and libraries for operations in AES transciphering.

cc @WoutLegiest @FlorentCLMichel

Filing this issue for tracking, beginners welcome here too. Will also be focusing on writing the rest of the AES kernels into a high level API friendly language to hook into this pass.