ShardingTaskExecutorPoolMaxQueueDepth was added in SERVER-91479 in order to limit how much work can build up on mongos under overload, but we didn't end up enabling it due to uncertainty about what the best value is. However, setting it to 1000 on the find_max_time_ms workload resolves the unavailability faced in the workload (before/after attached in comments).
We should figure out what the best value for this queue size is and then set this parameter, as bounded queues are key for the mongos availability story.