Describe the bug
The current decoder logic filters values that contain non-ascii characters. This works quite well, however, if the encoded value contains a unicode character (chinese symbols, ê, etc.) it will be silently skipped.

To Reproduce

1. Base64-encode a value that uses a valid UTF8 character
2. Attempt to decode it; it will be skipped.

Expected behavior
Segments with valid UTF8 characters ideally shouldn't be skipped. I don't know what the performance implications are of this.

cc @zricethezav @bplaxco