Partitioning adds query-plan complexity, but archiving old data becomes trivial. For a table with 100M+ rows and time-based access patterns, when do you commit to it?