Cover branches for databases that prefer unicode

[MathMagique]

A few code paths that handle ODBC's unicode columns are listed as not covered by our code coverage tool. The reason for this is that coverage information is collected on Linux and OSX only, and all databases they use run with prefer_unicode=False. The Windows builds do run tests with prefer_unicode=True, but they do not report coverage data.

Either have Windows builds report coverage data or let Linux/OSX builds run tests with prefer_unicode=True, which implies using an additional database.

[xhochy]

The simplest solution would be to also run MS SQL Tests on Linux. The MS SQL docker container makes this quite simple.