Add Firefox support

[Pehrsons]

See individual commits for more detailed comments.

[Pehrsons]

This comes from Orphis/webrtc-sandbox#19.
@alvestrand could you take a look?

[fippo]

Replacing all this stringsoup by https://github.com/fippo/simulcast-playground/blob/gh-pages/split-merge.js (which has worked with Firefox for years) would be better

[Pehrsons]

Ping @alvestrand

[jan-ivar]

This is phrased as getting Firefox working, but all looks like improvements that are to spec to me (e.g. avoiding unnecessary ssrc conflicts from mid->rid cut'n'paste swizzling). cc @henbos do you agree?

[fippo]

@jan-ivar according to some theories the spec forbids a=ssrc lines with simulcast. I am not a fan of that theory and could not even find spec language but Firefox including both mid/rid and ssrc is overspecifying things.

Your tone is not helping.

[fippo]

This is moving from mid/rid towards ssrc which IIUC is not what the spec (for better or worse) desires?

It turns out this shows what I believed to be a very interesting stats bug:

I couldn't repro on my playground (which uses the same GUM constraints) but chatting with @Orphis I figured out that the resolution of 1600x896 can be written as (1280+320)x(... wait this did not add up)
The GUM constraints are:

video: Object { width: 1280, height: 720, frameRate: "30" }

but

document.querySelectorAll('video').forEach(v => console.log(v.id, v.videoWidth, v.videoHeight))

shows the same numbers so it can not be a stats bug. track.getSettings returns height: 896, width: 1600

[Pehrsons]

This is moving from mid/rid towards ssrc which IIUC is not what the spec (for better or worse) desires?

We could also remove the ssrc lines but would then lose the rtx mapping, so should remove rtx as well.