Two fewer failures

[pbrisbin]

• Preserve newlines across tokenization, so we can catch invalid documents with unexpected newlines at parse-time

  I'm not sure if this was the right way to solve this, or if there's a way to let tokenization itself fail on unexpected newlines -- but not have the parser ever need to know about them. Let me know what you think.

• Replace calls to error with monadic failure, so we get parse errors, not exceptions, on certain invalid documents.

[clord]

tracking newlines into parser is useful, as they can be semantic.