k8s-access: use pkill instead of find to kill ssh process

Since we are reusing socket filename (=Config.id), launching multiple trackbone shell (consecutivelly, without cleaning ssh process between runs) prevents old k8s-access tunnels from being cleaned up. (we only list current inode)

Relying on pkill should also be ok on CI since jobs are running in different PID namespaces (so we shouldn't see other processes)