Implement specialization fast path for mixed types operations

[mratsim]

#49 and #50 will introduce overloaded proc to allow the following

• uint256 + uint64
• uint64 + uint256
• ...

Currently those will be implemented by converting the smaller int to uint256, meaning extra zero allocation.

As mentioned here, a fast (?) path can be use which avoids this allocation. It requires special casing/reimplementing most of the procs however to handle:

• carry in addition
• borrow in substraction
• carry in multiplication
• Division/mod I'm not sure, I don't think we can specialize, we need to shift the most significant bit of the divisor to the MSB of the dividend to do fast division.
• xor/or/and special iterator (or do result = input64.u256 and then result = result xor input256)

[zah]

Just a probably obvious tip: For +, *, xor/and/or, you can implement only one of those operations and remap the other variant with a template.