RTC-512 Add Room.State module and peer disconnected timeout

[Rados13]

This PR introduces two changes:

• extracts room state as a separate module and moves part of the logic related to modifying this state to that separate module
• adds support for peer disconnected timeout, after which, if peer doesn't connect, the peer will be automatically removed from the room
• add logs about WebSocket disconnects

Acknowledging the stipulations set forth:

• I hereby confirm that a Pull Request involving updates to the Software Development Kit (SDK) has been smoothly merged, currently awaits processing, or is otherwise deemed unnecessary in this context.
• I also affirm that another Pull Request, specifically addressing updates to the documentation body (commonly referred to as 'docs'), has either been successfully incorporated, is in the process of review, or is considered superfluous under the prevailing circumstances.

PR to docs
PR to elixir SDK
PR to Python SDK

[codecov]

Codecov Report

Merging #178 (1e308e6) into main (67bf454) will increase coverage by 0.36%.
The diff coverage is 87.61%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #178      +/-   ##
==========================================
+ Coverage   86.88%   87.24%   +0.36%     
==========================================
  Files          73       74       +1     
  Lines        1433     1482      +49     
==========================================
+ Hits         1245     1293      +48     
- Misses        188      189       +1     

Files	Coverage Δ
lib/jellyfish/peer.ex	85.71% <ø> (ø)
lib/jellyfish/room/config.ex	96.42% <100.00%> (+0.27%)	⬆️
lib/jellyfish/room_service.ex	87.01% <100.00%> (ø)
lib/jellyfish_web/api_spec/room.ex	100.00% <ø> (ø)
lib/jellyfish_web/controllers/room_controller.ex	100.00% <100.00%> (ø)
lib/jellyfish_web/peer_socket.ex	82.00% <71.42%> (-2.10%)	⬇️
lib/jellyfish/room.ex	83.10% <86.27%> (-0.17%)	⬇️
lib/jellyfish/room/state.ex	88.07% <88.07%> (ø)

... and 1 file with indirect coverage changes

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 67bf454...1e308e6. Read the comment docs.

[roznawsk]

I find the peerlessPurgeTimeout and peerDisconnectedTimeout a bit simmilar. How about renaming the peerlessPurgeTimeout to roomIdleTimeout?

[sgfn]

The room doesn't have to be idle for the purge to happen, though -- there can be components present...

[sgfn]

[nitpick] monotonic_time returns integer()

[sgfn]

I'm really unsure about the changes introduced here. I see why we're choosing to extract the state to a separate module, but as it stands, the implementation is "uneven", the abstraction level fluctuates -- e.g. notifications are sometimes emitted by the functions in state, sometimes by the functions here; calls to State can update a single key in the struct, or do a lot of additional logic as well.

IMO the State module should mainly be considered with, well, keeping track of the room state -- any and all additional tasks, such as emitting telemetry events and notifications, should be performed by Room. However, ensuring this distinction in the code could potentially diminish the benefits we achieved from splitting the state in the first place...

Discussion is welcome

[Rados13]

I agree that the abstraction level fluctuates.I thought about how to better handle situations like broadcasting events, but it is pretty hard because many events require a lot of information e.g :track_removed event the API of the remove_track would look terryifing to return all needed information. As I see currently I could move the last two events from Room to State, but I am not sure if this will help a lot.

[roznawsk]

In my opinion it's a good thing that we moved some of the logic to the State.
It is true however that it is sometimes unclear what actions take place in the State and which withing the Room.

I think we can clear this up a bit.

First, we can move sending all Events to State.
Maybe it would be nice to send them from Room . But since all of them but two are emitted from State we can just make it a convention.

We can also get rid of any changes to state struct within the Room and move it to the State.

[roznawsk]

LGTM 🎸