The domain of integral bounds given to `repeat_directive` is underspecified

Perhaps this should be both rejected in the code and described in the new documentation.

• The documentation should note that repeat operates on closed bounds.
• x4::repeat(negative) should be rejected
• x4::repat(negative, ...) and x4::repat(..., negative) should be rejected
• x4::repeat(max, min) should be rejected
• x4::repeat(0) should be rejected
  • This is semantically very vague.
  • As of now, I confirmed that it actually results in always succeeding parser (regardless of the underlying parser p given to repeat(0)[p]), similar to eps

Proposed solution:

• The basic concept should be changed from std::integral to std::unsigned_integral
• x4::repeat(max, min) should throw std::domain_error
• x4::repeat(0) should throw std::domain_error