Restructure connection checks to look for heartbeat timestamps

[third774]

This PR removes the websocket connection checks in the alarm, and instead looks for a heartbeat timestamp to evaluate whether or not to consider a user "timed out". This should hopefully resolve the issue we've seen for a while where sometimes random users are "timed out" incorrectly.

[nils-ohlmeier]

No major concerns with the change.

One bigger question though is: do we want to continue to base the decision if a participant is still in a given meeting or not based on the state of the WS alone?
I think this gets tricky (or is impossible) to look at the state of the users PeerConnection as well and react differently if the PeerConnection appears to be connected and alive, while the WS is in trouble. Maybe more of an idea for another future improvement.

[third774]

do we want to continue to base the decision if a participant is still in a given meeting or not based on the state of the WS alone?

This change removes that almost completely. We will now only remove a user if:

1. The user sends userLeft message
2. The state broadcast to a user fails (we might even remove this one TBH?)
3. The cleanup sees that the latest heartbeat is too old

What are your thoughts on this?

[nils-ohlmeier]

do we want to continue to base the decision if a participant is still in a given meeting or not based on the state of the WS alone?

What are your thoughts on this?

Let's give it a try. We can come back to this topic as needed. :-)