* When splitting, use rsplit and 1 just in case the string has more pipe
characters than anticipated
* When converting to text, make decoding errors an error instead of
silently corrupting data
* When converting to text for message strings, use the default error
handler as that is already surrogate_then_replace