Add monthly kWh load output to CAIRO runs for demand-flex load delta comparison

[alxsmith]

What

Add a new per-building monthly kWh output to run_scenario.py so that post-processing (build_master_bills) can compute elec_load_delta — the kWh difference between a base (non-flex) run and its demand-flex counterpart.

Today CAIRO computes aggregated monthly loads internally (agg_load from process_building_demand_by_period) but never writes them to disk. Only bills ($) and BAT metrics are saved. To compare kWh between the base run (elasticity=0.0) and the flex run (elasticity=-0.1), we need the monthly load per building from both runs.

Why

The ConEd hp_seasonalTOU_flex scenarios (runs 13-16) are the only runs with demand flexibility (elasticity=-0.1). Analysts need to see not just the bill impact but the actual kWh load shift per building/month to understand how much consumption moved between TOU periods. Without this, the only record of the shift is the demand_flex_elasticity_tracker.csv (per-building epsilon values), which doesn't give monthly kWh totals.

How

1. Save monthly kWh per building during run_scenario.py: After bs.simulate() completes, the agg_load DataFrame (output of _vectorized_process_building_demand_by_period) contains grid_cons and load_data by (bldg_id, month, period, tier). Sum grid_cons by (bldg_id, month) and write to bills/elec_load_monthly.csv (or similar) alongside the existing bill CSVs.
2. Add elec_load_delta to build_master_bills.py: For flex run pairs (e.g., 15+16), accept a --base-run-delivery / --base-run-supply argument pointing to the non-flex counterpart (e.g., 3+4). Read elec_load_monthly.csv from both, compute delta = flex - base, and join into the master bills table. For non-flex runs, elec_load_delta = 0.
3. Run pairing for ConEd: base run 1 ↔ flex run 13, 2 ↔ 14, 3 ↔ 15, 4 ↔ 16.

Deliverables

• run_scenario.py saves elec_load_monthly.csv per CAIRO run (bldg_id, month, grid_cons_kwh)
• build_master_bills.py accepts optional base-run args and adds elec_load_delta column
• elec_load_delta is 0 for non-flex runs, non-zero for flex runs
• Verified on ConEd runs 15+16 vs 3+4

[linear]

RDP-178