Implement UInt128 with only BigInteger256

[weikengchen]

The current code adds the implementation of UInt128, which is the constraint version of the primitive type u128.

The current implementation uses num-bigint instead of BigInteger256. However, there is no reason not to use BigInteger256 beside the (slight) code complexity. It should be possible to revise the current implementation to be based on solely BigInteger256.

There is a significant benefit of not using num-bigint because r1cs-std is a fundamental crate that is used by many applications. It is good to just keep it lightweight.

[Pratyush]

On the other hand, if nonnative is merged into r1cs-std in the future, then we'll have the num-bigint dependency anyway