Fixed loading profile with non-ASCII path on Windows

[radarhere]

Resolves #4897 - in the same way that #3785 resolved #3145, by loading data from a non-Unicode path on Windows into memory.

Also corrects a message in the test suite from that PR.

[hugovk]

What does "non-Unicode" mean here? ASCII?

[fireattack]

Yeah, was about to comment the title before :p I think he meant non-ASCII.

[nulano]

LCMS2 calls fopen internally, same as FreeType, which on Windows expects a string encoded with the current ANSI codepage: https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/fopen-wfopen?view=vs-2019#remarks

Python returns strings in the utf-8 encoding for the s parameter in PyArg_ParseTuple. These two encodings are only compatible in the ASCII range unless the current codepage is UTF-8, but that is not yet widely supported.

So it seems non-ASCII is the intended message.

[hugovk]

Thanks!