Speed up encoding

[andrewthad]

It should be possible to speed up encoding by doing a machine-word-at-a-time walk of strings, checking if they have any characters that need to be escaped. This includes:

• Everything < 0x20
• 0x22 (double quote)
• Everything >= 0x7F

Surely there's a bit twiddle hack somewhere for this. I've not found it yet, but I'll come up with it if I have to.

If there are no characters that need to be escaped, then memcpy can be used to copy.

[kozross]

There are twiddling hacks to do a range match, but I suspect that having to combine two ranges plus a single match won't benefit you without SIMD. The hacks are here.