SelectBlobs: Replace Athena with pyspark.sql
The growing dataset size means we now hit Athena's 30 min timeout. Using pyspark takes longer (3h) but spares the slow download from S3 and does not have a timeout.
This adds a dependency on pyspark, which is a 300MB package full of JARs; but we might use it for other workloads in the future.
Edited by vlorentz