speech recognition API

Here's an update to the speech recognition API proposal. 
I added some discussion based on last week's call -- as we discussed, all
parameters are set on the recognizer object, not as part of the call to
start recognition. There's also some discussion of the issues that I think
are still up in the air -- whether there should be a way to set multiple
parameters in the same call, and what it means to be able to control which
portions of the captured audio
are sent to the recognizer.
I also added more details on the API calls that were in the earlier
proposal.