Summary
See #2976 for background. That PR adds support for non-ASCII features back to the C++ and Python libraries. A test in the R package was added in that PR but was breaking for reasons that weren't obvious, see for example #2976 (comment).
I think that closing this issue will mean changing lgb.encode.char(). You'll know that a f ix is working if you remove the testthat::skip() call added in #2976 .