`char_(from, to)` and `char_("set")` should reject invalid code units

This issue is not just about the implementation quality, it is actually about the validity of the design.

There are currently multiple layers of misfeature:

1. char_('Z', 'A') -- should be rejected
2. char_(from, to) with non-single code unit param -- should be rejected
3. char_("Z-A") -- should be rejected
4. char_("あ") -- should be rejected, while char_(U'あ') should be maintained
5. char_("abcあdef") -- should be rejected, while char_(U"abcあdef") should be maintained

The real problem is that we can fix x4::unicode but can't do anything about x4::standard and x4::standard_wide.

Non-unicode variants operate on the execution character set, and moreover, our implementation currently uses <cctype>, which utilizes the C locale. So they don't have well-defined meaning for "character ranges" or "character sets".

Note that I used the term "code units", which is the Unicode terminology, merely for explanation, and C locales may or may not be Unicode in practice. Even if some C locale happens to be Unicode, the case like (5.) cannot be resolved for x4::standard because multi-code-unit character like u8"あ" is not representable by the character parser; i.e., (char)u8'あ' is invalid even when char is UTF-8.