Feature: controller: allow full interval specs for relevant cluster properties

Previously, cluster options that took a duration (dc-deadtime,
shutdown-escalation, cluster-recheck-interval, transition-delay,
join-integration-timeout, and join-finalization-timeout) used crm_get_msec().

Now, they use crm_parse_interval_spec(), so they can additionally take ISO 8601
durations.

This involves changing their storage from int to guint (which is actually the
main goal, to be consistent with how they're used with g_timeout_add()).