CPTokeniser should be able to work on a stream of input rather than the whole NSString

[beelsebob]

This would allow it to begin tokenising mid way through downloading content, rather than requiring it all to be in memory before beginning.