Use read_utf8() and write_utf8()

[krlmlr]

from xfun, which is a dependency anyway via rmarkdown -> knitr. This fixes encoding problems on non-Latin locales.

Was meant to address #234, but can't fix it unfortunately.

[jennybc]

What's the status of this PR @krlmlr? Is "WIP" still accurate or should it be considered "ready to go"?

[krlmlr]

I think using read_utf8() etc. is better than the current implementation, but it won't fix the bug that triggered this PR.

Removed the AppVeyor cache to fix the broken build, also turned on binary installation. Let's see what the Windows test shows.

[krlmlr]

I guess we're good to go here.

[jennybc]

OK I will experiment with this personally over on Windows.

[jennybc]

I can imagine we'll need refinements. And I'd like to add tests that means something. But this seems like (the first) step in the right direction. Merging.

[krlmlr]

Thanks. I don't think the caching on AppVeyor is currently worth its price, especially if installation from binary is enabled. If problems occur again, we can cherry-pick 03c0d5a.