If we replace the lisp-to-JSON encoder with native code, we would not need the jansson library for it and it would be faster.

There is ongoing work on a JSON-to-lisp parser, but the author has made it clear that he does not have time to write an encoder, so I spent a morning mashing up the attached patch.

It generally produces the same result as the old code, except:

- The old code incorrectly accepted strings with non-Unicode characters (raw bytes). There is no reason to do this; JSON is UTF-8 only.
- The old code spent a lot of time ensuring that object keys were unique. The new code doesn't: it's a garbage-in garbage-out type of situation.

The new code could do with some optimisation but it's already about twice as fast as the old code, sometimes more.

I'd be very happy if someone could test it with packages that use this interface (json-serialise, json-insert).