Make StereoPannerNode a-rate and spec the algorithm

[padenot]

It was somewhat obvious, but it's clearer to also have the formulas for [-1, +1]. Also I made it a a-rate mostly because it's technically just two gain nodes with carefully computed values, and gain nodes are a-rate.

This fixes #439.

[hoch]

@padenot
I think the algorithm on the change is incorrect. We still need to take care of the polarity of pan (azimuth) value in stereo->stereo case. I believe it should be same as equal power panning process in the original panner node.

// normalization for mono->stereo
x = (pan + 1) / 2;

// for stereo->stereo
if (pan <= 0) {
  x = pan + 1;
} else {
  x = pan;
}

x=pan doesn't mean anything particular, but I think it is better to keep it for the reason behind.

[padenot]

Ha, good catch. In fact, what you say is exactly how I implemented it in Gecko, it's just me writing spec text too late, heh :-)