[FIX] Memory leak generated by Stream Cast usage

[rodrigok]

Proposed changes (including videos or screenshots)

Stream Cast uses a different approach to broadcast data to the instances, it uses the DDP subscription method that requires a collection on the other side, if no collection exists with the given name broadcast-stream it caches in memory waiting for the collection to be set later. The cache is cleared only when a reconnection happens.

This PR overrides the function that processes the data for that specific connection, preventing the cache and everything else to be processed since we already have our low-level listener to process the data.

More info

Bellow an example of the memory usage of a deployment using the normal matrix event broadcast among instances during the first 9 days and then using the Stream Cast to broadcast events during the last 5 days (with a restart every day due to the memory usage).

[sampaiodiego]

doing this actually breaks streamcast, all events are ignored and communication doesn't happen.

another way of preventing this leak to happen, is to actually register a store for the broadcast-stream collection.

so I changed the current line 180 (connection._stream.on('message', function(raw_msg) { for the following code:

	connection.registerStore('broadcast-stream', {
		update({ fields }) {
			const { streamName, eventName, args } = fields;

			if (!streamName || !eventName || !args) {
				return;
			}

			if (connection.broadcastAuth !== true) {
				return 'not-authorized';
			}

			const instance = StreamerCentral.instances[streamName];
			if (!instance) {
				return 'stream-not-exists';
			}

			if (instance.serverOnly) {
				return instance.__emit(eventName, ...args);
			}
			return instance._emit(eventName, args);
		},
	});

looks like even a less hacky solution, wdyt?