Avoid intermediate bytestring construction when encoding

[meiersi]

Currently, lot's of intermediate bytestrings are constructed during encoding. This is unnecessary and could be avoided by adding a toFieldBuilder method to the ToRecord typeclass. I expect at least a factor two speedup in encoding from this change.

[tibbe]

I've thought about doing that and might do it eventually. Right my whole encoding implementation is very naive (have a look at the encode function) so I expect there's even more easily wins to be had. Since my implementation is already faster than the Python csv library (which is written in C) I didn't bother so far.