Just connecting two of your stacks' switches without configuring a LAG would result in a network loop which would be kept disconnected by Spanning Tree (i.e. one of your links would be constantly held in a "blocked" state). In the case of link or switch failure, the failover times for STP are approximately 50 seconds, RSTP would be considerably faster around 1 second.
Configuring a LAG would result in a trunk where both links would be held active and traffic would be distributed among them depending on the algorithm chosen. The failover / link failure notification time would depend on the implementation but typically is within the magnitude of 1 second.
If you do not need the additional bandwidth brought by Link Aggregation, it is mainly a question of which implementation you trust more to handle all of your possible failure cases gracefully. You might want to subsequently set up both configurations and test a couple of failure scenarios and see if the failovers work as expected and the times suit your needs. Some examples of what you could do:
- switching off a stacked switch connected to the uplink
- pulling one of the uplink cables
- breaking (or over-bending) the glass fiber of one of the uplink cables (preferably two times - once for each direction's fiber)
- completely separating out a single switch connected to one of the uplinks out of your stack by disconnecting both stack links