long reconnect times

[iameli]

With catalyst multi-node, disconnecting an RTMP stream takes a long time afterward. This happens because incoming connections check against the load balancer, notice that the stream already exists, and exit.

What we should do instead:

• When PUSH_REWRITE fires, we attach a random id to the incoming stream. Stream abc123 gets rewritten to e.g. video+abc123-ffffff.
• Catalyst cluster members publish their current streams into the kappa state.
• When nodes want to handle playback, they query that kappa state, discover which stream has the address at which region, and victory.