WIP: Allow different string escaping strategies

[PaulGrandperrin]

This is a WIP trying to solve the problems exposed in Issue #51, PR #58 and PR #65.

TL;DR;
When serializing to JSON, some people might want:

• Only escaping characters as mandated by ECMA-404, with maximum performance.
• All Unicode control characters to be escaped (C0, DEL and C1), just like in Python and Ruby, with slightly degraded performance.
• All non-ASCII characters to be escaped, and therefore generating ASCII compatible JSONs, juste like in Python with the option ensure_ascii =True, with even more degraded performance.

This PR tries to address all these needs.

[oli-obk]

The travis failure is legit:

src/lib.rs:119:1: 119:21 error: #[feature] may not be used on the stable release channel
src/lib.rs:119 #![feature(unicode)]

and on nightly:

src/ser.rs:650:36: 650:48 error: this function takes 0 parameters but 1 parameter was supplied [E0061]
src/ser.rs:650                         match char.encode_utf16(&mut utf16).unwrap() {
                                                  ^~~~~~~~~~~~
src/ser.rs:650:36: 650:48 help: run `rustc --explain E0061` to see a detailed explanation
src/ser.rs:650:61: 650:67 error: no method named `unwrap` found for type `core::char::EncodeUtf16` in the current scope
src/ser.rs:650                         match char.encode_utf16(&mut utf16).unwrap() {

[dtolnay]

Closing due to inactivity. I would still be interested in offering a choice of what characters to escape if anyone can find a nice way to do it.