Use mesh peer count to determine required subnets

Currently, we use the number of connected peers on a subnet to decide on whether to connect (and potentially discover) more peers. This is problematic, as we might have e.g. no mesh peers, so we should try to find more peers to get a mesh.