Change RegExps to work on 16-bit wide characters on all platforms. Removes wchar_t refs

[tcr]

This removes the use of "mbtowcs", replacing it with an explicit UCS-2 function and and hsregex patch that fixes its characters to 16-bits wide.

[tcr]

Will dig through the comments and update this patch. (Github's interface is really confusing me with its comments UX today.)

[tcr]

@natevw Updated. To recap, hsregex is blindly using 16-bit wide code points and so we can seamlessly get away with UTF16, because we'll never have a non-surrogate pair originating from a colony-internal string representation. (Hence the function name tm_str_to_utf16.)

Also I printed out your string diagram and pasted it on our wall.

[natevw]

Looking much better, although I think you should change the signature (and relevant internals) to const uint16_t ** const dstptr — this not only will turn *((uint16_t *) (buf + buf_pos)) = uchar; into a much clearer buf[buf_pos] = (uint16_t) uchar; but makes it clear to callers that it is native endian (i.e. a chunk of 16-bit ints) rather than a ??? endian chunk of bytes.

[tcr]

@natevw I've changed the function definition to take a tm_endian_t argument, as the buffer methods do. Hopefully this will be useful with the Buffer encoding options as well.

[natevw]

@tcr That works, sorry for not reviewing sooner. R+ assuming it actually works in practice too ;-)