All unit tests written except chain and chainl. Removed left_factor_action, we'll handle left-recursion with our caching strategy.

Fixed inconsistent signature of charset parser, registered (still stubbed) unit tests so the build doesn't break :)

Conflicts:
	NOTES

Fixed build (completed butnot combinator). Also, fixed bug where sequence wasn't bailing out if one parser failed.

All sorts of makefile hacks, plus a new fastpath for the bitreader 

Parsers for sequence and choice are done. Got rid of join_action, we didn't need it. parse_result_t now contains a single parsed_token_t (and whatever we decide to jam into it later), so that sequence makes sense.

 * token: matches a sequence of bytes (with length)
 * ch: matches a single byte
 * range: matches any byte within the range [lower, upper] (inclusive)
 * join_action: joins the results of another parser with a separator
 * negate: matches the opposite of any single-character parser
 * end_p: succeeds if there's no input left to parse
 * nothing_p: always fails

One other big change: the AST is now a GSequence of parsed_token_t's. WARNING: This is not actually enforced, because C.

Also tweaked the makefile a little (which will get clobbered in TQ's next commit) and added some documentary comments to hammer.h.