Tests no force ws; fix failure to callback on presence event causing implicit attach

[SimonWoolf]

Started out as a PR to just stop forcing web_sockets when not necessary (fixes #127), ended up also fixing a couple bugs that that exposed:

• If there's an error in an implicit attach caused by sending a presence message, callback with that err. (was only a problem with presence not with normal messages since normal messages get their callback called in RealtimeChannel#failPendingMessages). + test (and another test for normal messages, though that one was already passing)
• Fixed an onNack callback typo in protocol.js. (indicator that codepath for NACKs is not well-tested?)

[paddybyers]

Thanks