Improve performance of length on UTF8String

[ScottPJones]

I rewrote the u8_charnum function to improve the speed of getting the length of a UTF-8 string.
(in my tests, it was approximately twice as fast as using u8_charnum, and many times faster than
the old approach using u8_strlen [which should be removed, along with u8_strwidth, because if called with a malformed string without a terminating \0, they can overrun the buffer, potentially causing an access violation])

[JeffBezanson]

Awesome, thanks!

[stevengj]

LGTM.

[stevengj]

As an aside, I noticed a bunch of places in string.jl that test length(s) == 1 and similar, which is O(n) when it should be O(1). We should really have a fast islength1(s) routine or something like that—should only be a couple lines long.