Fix "�� is not JSON serializable" on Python 3

[hugovk]

Fixes #258.

[JustAnotherArchivist]

I just came across this and am wondering what the error handler string "~E~" is. I think that would just throw a LookupError when any encoding error occurs (cf. '\ud800'.encode('utf-8', '~E~') and codecs.lookup_error('~E~')). Is that intentional? Why not just strict, which raises a UnicodeEncodeError instead? (Yeah, I realise this was two years ago.)

[hugovk]

It was probably taken from https://stackoverflow.com/a/46202119/724176:

in this example, this will replace invalid characters/data with the string "~E~"

But I also see this PR didn't add any test case with that string, so we could certainly improve that and re-evaluate this!

[JustAnotherArchivist]

Yeah, about that...

import ujson

class A:
	def __repr__(self):
		return '\ud800'

ujson.dumps(A())

→ Segmentation fault

I'm not sure whether that replacement was ever a thing (couldn't find it anywhere in the Python docs, not even back in Python 3.0), but it definitely isn't now. I guess that belongs in the 'wrong Stack Overflow answers' cabinet.

This is similar to #522, and actually I realise now that there are more issues with that string conversion as well. Will follow up with another PR.

[hugovk]

Thanks!