Allow 'single-character' mi's to include trailing combining characters (mathjax/MathJax#2617)

[dpvc]

This PR doesn't sound combining characters when counting the length of the content of an mi element in order to determine the mathvariant to use. This allows things like \not a to render as an italic "a" rather than an upright one.

Resolves issue mathjax/MathJax#2617.

[zorkow]

I note that you are making provisions for characters with combiners >1000.
I never got those to work. E.g., '\u0041\u0300'.normalize() works fine, however, '\u0041\u1ab0'.normalize() still gives me two characters. Am I doing something wrong?

[dpvc]

'\u0041\u0300'.normalize() works fine, however, '\u0041\u1ab0'.normalize() still gives me two characters. Am I doing something wrong?

No, that is what should happen. '\u0041\u0300'.normalize() gives you one character because there is a Unicode point U+00C0 for CAPITAL LETTER A WITH GRAVE, so the two can be combined into a single Unicode character. But U+1AB0 is the COMBINING DOUBLE CIRCUMFLEX ABOVE, but there is no single-character glyph for A with two circumflex accents, so this has to stay two characters. The point is that we want to count that as a single character, even though it is technically two, for the purposes of deciding whether it should be mathvariant="italic" or mathvariant="normal". The combining characters shouldn't make it count as two (causing it to become upright instead of italics), as it is semantically a single accented character, just like your first example.