Fixes: #2351

Problem

When a client makes N concurrent sync requests to pathfinder over the same sync protocol:

• N>3 pathfinder responds but pending connections from other peers are held up
• N>7 pathfinder initially responds but then just stops responding to the first client and does not react to other peers trying to connect

The mechanism causing this issue

It turns out I was wrong the first time I approached this issue and wrongly accused SelectAll in swarm's connection pool for the stalling. The fact that SelectAll stalls in that connection pool is only a symptom of processing slowing down elsewhere, this is what actually happens:

• a sync request is received, p2p_stream emits InboundRequest event, which is caught in the main loop and then re-emited outside the main loop as Inbound*SyncRequest through this channel
• this Inbound*SyncRequest event is taken from the channel here and a proper sync handler is called
• sync handler waits till DB finishes, so the event channel becomes full quickly
• backpressure through the event channel is exerted on the main loop, as the channel fills up and some of the next attempts to forward an event from the main loop results in waiting on a full channel here
• polling swarm for newer events cannot proceed, we hang on swarm.next()
• swarms' internal event queues for each connection fill up, SelectAll cannot move forward, it's not being polled fast enough
• swarm becomes unresponsive to external events

Config changes

The default value of max_concurrent_streams is back to 100.

Tests performed

Snapshot: sepolia
Number of clients: 10