cmd/devp2p: fix ping/pong race in discv4 tests

[holiman]

Nethermind fails tests spuriously on Hive. They fail basic ping test, because they send a PING not a PONG back. This is most likely a data race. When they get a PING, they're supposed to send both a PONG and a PING. However, the test framework assumes that the PONG will be the first response packet to arrive.

This PR modifies the post-PING-send expectations to both be laxer and stricter: it doesn't care what order the packets arrive, but also verifies that exactly one PING and one PONG is returned.

Tested against my local node:

$ ./devp2p discv4 test --run Ping --remote "enode://cc44695159f742595bf2fd9a9da13535733836c71cd5ca1179d76c11b95565f37c8014db177e90c9d7907b3a17439b3f51949c2144adbe4fef0e615d11900ca8@127.0.0.1:30303?discport=0"
-- OK Ping/Basic (1ms)
-- OK Ping/WrongTo (700µs)
-- OK Ping/WrongFrom (800µs)
-- OK Ping/ExtraData (700µs)
-- OK Ping/ExtraDataWrongFrom (700µs)
-- OK Ping/PastExpiration (300.9ms)
-- OK Ping/WrongPacketType (302ms)
-- OK Ping/BondThenPingWithWrongFrom (5.1ms)
8/8 tests passed.