Allow unknown characters to use operator table to determine class and node type. (mathjax/MathJax#3203)

[dpvc]

MathJax's handling of characters that aren't mapped to special functions (like ^ and _), or that match the variable or number patterns, is controlled by the RANGES and OPTABLE in the OperatorDictionary.ts file. When an unmapped character is encountered, the RANGES array is searched for a block in which the unknown character is found, and that gives the TeX class and MathML node type to use for the character.

These blocks are just a rough breakdown of the unicode ranges into groups that should be treated similarly, but it is not perfect, and some characters are mischaracterized by it. Two such are U+00D7 (×) and U+00F7 (÷), which are found in a block of Latin letters, which MathJax marks as TeX class ORD using an mi node. These two should be treated as operators, but they have not been singled out from among the Latin letters, so get the wrong MathML node type.

One solution would be to make the RANGES table more granular, but another is to note that U+00D7 and U+00F7 are present in the operator dictionary, so Mathjax could use that to find the TeX class and wrap it in an mo. The latter is implemented here. That means that any character that is in the operator dictionary will be put into an mo automatically, rather than use the RANGES table. That makes it easier to customize the operators recognized by MathJax without having to adjust the RANGES table.

Resolves issue mathjax/MathJax#3203.

[zorkow]

lgtm.