Phase 2: Turn the notebook pipeline into a real batch job

[ChantelRM]

Suggested pace: ~2 sessions this week

• Extract the core pipeline logic (Sections 0-6 of the notebook) into a script that runs non-interactively via Rscript
• Swap flat-CSV outputs (unified_ledger.csv, pursuit_rankings.csv, etc.) for DuckDB tables as the actual storage layer -- reuses the DuckDB dependency already in place for querychat
• Confirm the dashboard's app.R can read from the DuckDB tables instead of the CSVs, or documents clearly why it still reads CSVs if that's the final call
• Update README to describe the pipeline as a script + database, not just a notebook