Upgrade test started failing because older versions of redpanda don't
support the partition_autobalancing_mode flag (that was added to default
config instead of the deprecated enable_auto_rebalance_on_node_add flag.
Looks like the only tests that need a non-default value are
node_operations_fuzzy_test (that sets it explicitly) and scaling_up_test
(where we explicitly set the value)..

Sometimes in cithe leader table doesn't get updated in a timely fashion,
leading to sporadic partition_balancer_test failures, so looks like the
timeout of 30 seconds needs to be increased.