This improves performance on the tweet data set by between 10% and 50%, depending on the input.
We can now convert most types deriving Data. Also, change the parsing code from being based on Applicative to using an efficient continuation-based Parser type, with a Result type signalling the result.
We were on par with the json package in some cases, and slower in others. Now, with plain text, we're 25% faster, and for Unicode-heavy text, twice as fast.