chore: add metrics to redis collab stream

[Horusiath]

This PR comes with 2 changes:

1. CollabStreamMetrics type which allows us to track how many streams are we scheduling to be read at given time:

• collab_stream_reads_enqueued is incremented for every stream that has been observed. When the Redis xread returns and stream needs to be rescheduled for the next poll iteration, it also becomes enqueued.
• collab_stream_reads_dequeued is incremented for every stream that scheduler returned for xread. These two metrics should be fairly close to each other. If they reads_enqueued grows faster than reads_dequeued it means, that we schedule new stream read requests faster than we can handle them.

2. It fixes the issue when an observer was not correctly dropped and stayed inside of stream router for as long as no new objects where pulled from its corresponding stream.

[Horusiath]

The problem was that we only unscheduled the stream subscriber if it was closed AND stream returned at least one message that we wanted to forward into (now closed) subscriber. If no new messages from stream were received, it will be rescheduled again even if subscriber was closed.

This could mean that if we had 5000 subscribers in the past, but no new data was coming from their Redis streams, and only 5 of them where active atm. we still tried to XREAD 5000 streams instead of 5.