Unicode correctness

[maiself]

There are still some Unicode conversion problems in the adapter. Most should be easy to fix (std::string -> std::wstring), but they need to be located. Tracking them here may be helpful. Tests to ensure things continue to behave correctly would also be good.

There is a particular issue with writing unicode characters to sys.stdout / sys.stderr. Normally these are initialized as having a utf-8 encoding, but in pyjs they are initialized as ascii. This seems to be a result of how pybind11 initializes the scoped interpreter. I'm not sure which project should fix this or what the best way to do so is.

There may also be other yet to be discovered unicode issues.