Too small MaxGossipDelay results in disconnecting from all peers in testground

[evan-forbes]

If we use the default parameters using the v2 mempool in testground tests results in the tests not finishing. After increasing the MaxGossipDelay (to 20-30 seconds, although lower values have not been tried and could work), the tests reliably finish. More information is needed to properly debug this, however looking at the logs we see that nodes are unable to find peers. When this occurs, the tests often reach a low height and stops being able to reach consensus.

More data should be collected to determine the exact cause of such a bug, specifically data around peering. atm, testground doesn't have access to the standard metrics so we've resorted to using the tracer.

[cmwaters]

So just from looking at the code, there doesn't seem to be any path which would cause the node to disconnect with their peer. There are three places that this currently happens:

1. If the WantTx has a txhash that is the incorrect size
2. If SeenTx has a txhash that is the incorrect size
3. If the peers sends a message that is not a SeenTx, WantTx, or Tx

If we can't see any of these errors then I'm not sure how the CAT pool is disconnecting from peers (unless because of some timeout). I also can't explain why extending the timeout solves the problem