The GitLab implementation had a couple places where we made wrong

assumptions about header types. This wasn't an issue on Review Board 3,

but on 4 we're more strict about types, and this resulted in some errors

when trying to determine the API version.
The API token was being turned into a byte string before being put into

a header. This was the correct type on Python 2, but not on Python 3.

The type checking caught this, but we didn't have unit tests that ran

through this code.
The logic being used here was redundant anyway.
We now set the header using a Unicode string (which is what our HTTP

requests now expect), and we use _get_private_token() to decrypt the

token instead of duplicating logic. On top of this, our processing of

the Link header for pagination now looks up and converts native

strings instead of assuming Unicode strings.

Reproduced the issue in bug #4928 in tests and verified the fix on

Python 2 and 3.