Remove feed()

[tjvr]

Implementing a streaming API on top of a RegExp-based tokenizer is fraught with difficulty, as discussed in #36. It's not clear what to do when a token overlaps with a chunk boundary; should we:

1. Require users to feed chunks which are already split on token boundaries
2. Buffer input until we get a regex match that doesn't extend to the end of the buffer —which introduces unfortunate restrictions on the lexer definition.
3. Re-lex the entire input when we receive new data, and somehow notify the consumer to discard some (or all!) of the old tokens (ouch)
4. Throw everything away and write our own RegExp implementation from scratch; then we can directly query the FSM to see if the node we're on can transition past the next character!

Only (1) seems like a workable solution. Excepting (4) (!), the rest you can re-implement yourself on top of reset(). So I propose removing feed() and re-introducing remaining() [which would return buffer.slice(index)]. Thus Moo is a fast lexer core; people can extend it with hacky strategies as necessary. :-)

[tjvr]

@Hardmath123