Incorrect results for string functions on args with higher-plane characters

Some functions are improperly treating strings that contain higher-plane Unicode characters (i.e., those with code points at U+10000 and higher, including most emoji) as if each of those characters were two characters long and unintelligible.

Split("🦓🦊🐺𐊀","")
→
Table({Value:"�"},{Value:"�"},{Value:"�"},{Value:"�"},{Value:"�"},{Value:"�"},{Value:"�"},{Value:"�"})
(Should be Table({Value:"🦓"},{Value:"🦊"},{Value:"🐺"},{Value:"𐊀"}))

Left("xyz🅰🅱🅲", 6)
→
"xyz🅰�" (note the ending character which is apparently half of the 🅱 emoji; this should just evaluate to the same string as was passed in)

Similarly, Right("🅰🅱🅲def", 6) → "�🅲def", and Mid("🅰🅱🅲def", 2, 4) → "�🅱�".

I believe something is going awry with how PowerFx is handling characters wider than 16 bits, and strings aren't being kept in a consistent translation format, which is leading to these errors.

Start by reproducing the Split, Left, Right, and Mid examples from the issue with higher-plane Unicode characters. Trace the string handling used by these functions and add coverage for the shown cases; done means each function preserves complete Unicode characters without replacement characters or split surrogate pairs.