Fix some long-standing UTF-8 bugs.

We we incorrectly rejecting U+fffe and U+ffff, and incorrectly accepting
characters above U+10ffff (see https://tools.ietf.org/html/rfc3629
section 12 for that restriction).

Bug: http://lists.landley.net/pipermail/toybox-landley.net/2017-September/009146.html
Test: ran tests
Test: also ran the exhaustive test from that email thread
Change-Id: I8ae8e41cef01b02933bd4f653ee07791932b79a5