Preserve integer sizes in NumberLiteral#int_bin_op

[collidedscope]

Presently, the following macro raises a very surprising Error: Arithmetic overflow:

{{ 3000000000u64 % 2 }}

This is because NumberLiteral#int_bin_op converts the operands to Int32s via #to_i, but this unnecessarily and incorrectly discards information about the kinds of the operands. Since by this point we've eliminated the possibility that either of the operands are floats, we can safely cast them to Int in order to preserve their sizes.

[jhass]

Good change! Can you look into adding a spec?

[straight-shoota]

Perhaps we could refactor the entire method to better utilize type safety?

    def int_bin_op(op, args)
      NumberLiteral.new(bin_op(op, args) do |me, other|
        if me.is_a?(Int) && other.is_a?(Int)
          yield me, other
        elsif me.is_a?(Float)
          raise "undefined method '#{op}' for float literal: #{self}"
        else
          raise "argument to NumberLiteral##{op} can't be float literal: #{self}"
        end
      end)
    end

[straight-shoota]

I've added a spec and refactored the implementation of the method.