Handle encoding and payloading better in webrtc

[Manishearth]

Currently the webrtc implementation assumes that things are supposed to be opus/vp8, and is rather inflexible with encoding.

Furthermore, we don't yet do payloading (decodebin auto-depayloads).

We should be assuming input streams are raw, and the webrtc code (not the streams code) should be running them through the appropriate encoder and payloader before hooking them up.

(This will mean that we will need to stop doing a caps intersection for sink pad selection and start doing a media type match, which is nicer anyway)