peer: add data channel worker

[daniel-abramov]

This relates to #133

The effect of this change is not as large as anticipated though. It turns out that for smaller messages in conferences with up to 6 people, the operation of sending to the data channel on the peer level took no more than 20µs on average. The peaks occur from time to time (up to 1 ms), but they are not as large, so the effect of this worker is not really that noticeable.

So now I question if it's worth it at all. Decided to file a PR to at least demonstrate how simple and small the change was and where it was done. We can close if we think that the complexity does not justify the improvement.

Please check the details: #133 (comment)

[dbkr]

Looks good: we can always try it & see if it makes a difference and back it out if it doesn't seem worthwhile.

[daniel-abramov]

Closing this as obsolete as it turned out that it does not bring any significant performance, yet it complicates error propagation, so it seems like it's not worth it (the pros don't outweigh cons unfortunately).