Clarify spark.default.parallelism

It's the task count across the cluster, not per worker, per machine, per core, or anything else.