[mlir][arith] Optimize arith.cmpi based on integer range analysis.

Add a pass which do arith dialect ops optimization based on integer range analysis (only cmpi for now).

Differential Revision: https://reviews.llvm.org/D140629