Server version checks are overly pessimistic

When a cluster of three MariaDB 10.11 servers is used with MaxScale, the lowest common server version is 10.11. If one of the instances is down when MaxScale starts up, the version of that node is unknown and is marked as zero. This causes MaxScale to assume that only the very basic network protocol is supported as in the version comparison the version of the node that is down is zero. This means that some of the more recent features are disabled even if all of the other nodes that are up would support it.

An improvement to this would be to detect the protocol version difference of "unknown" servers later on during the connection creation phase where the very unlikely protocol mismatch can be handled by closing the connection. This is less efficient than simply not connecting but the supported protocol versions of a server that cannot be reached cannot also be known in advance and thus the mismatch can only be detected after-the-fact. For 23.08, this simply means igoring servers whose versions are unknown in the version resolution phase.