Remove fancy-regex dependency

[reese]

Currently, any time we convert a string to use double quotes, we compile and run a regex for every string. Benchmarking the large test fixtures, Regex::new alone takes up more than 2% of the overall runtime, with single_to_double_quoted taking up 3.5%. We could cache these or precompile the most common ones, but honestly using a regex here is overkill -- we're basically just checking for slashes followed by one of the delimiters, which we could do by iterating over the characters instead.

Replacing the fancy-regex usage with a manual implementation shows single_to_double_quoted going from 3.5% -> 0.5% in my benchmarking, plus it's one fewer dependency.

(This is also part of #832, since fancy-regex takes UTF-8 strings.)

[froydnj]

I assume we have good test coverage for this?

[reese]

Yeah, this should be covered pretty thoroughly by the string literals stress test