Add demand-flex analytical validation and elasticity calibration

[alxsmith]

Implements the demand-flex elasticity calibration and validation framework for NY HP rate design. The core analytical question: what elasticity parameter makes our constant-elasticity load-shift model reproduce the aggregate demand response observed in real-world TOU pricing pilots?

What this PR adds

Calibration (utils/pre/calibrate_demand_flex_elasticity.py) — Sweeps candidate elasticity values for each NY utility and finds the one whose peak reduction % matches the Arcturus 2.0 empirical target. Computes both "no enabling technology" and "with enabling technology" recommendations using the corrected Arcturus regression (single model with interaction term: no-tech slope = -0.065, with-tech slope = -0.111 = -0.065 + -0.046 interaction). Writes both sets of seasonal elasticities to config/periods/{utility}.yaml under elasticity and elasticity_with_tech. Sweep range is configurable via --epsilon-start/--epsilon-end/--epsilon-step (default -0.04 to -0.50, step -0.02). Can compare analytical savings estimates against a completed CAIRO batch (--compare-batch, --with-tech).

Seasonal elasticity runtime (utils/cairo.py, utils/demand_flex.py, run_scenario.py) — The shifting pipeline now accepts a {season: epsilon} dict in addition to a scalar, resolving the per-season value inside _shift_season. This lets summer and winter have independently calibrated epsilons matching their respective price ratios.

Scenario YAML automation (utils/pre/create_scenario_yamls.py) — Reads enabling_tech from the Google Sheet and selects elasticity_with_tech from the periods YAML when empty or true (default), elasticity only when explicitly false/no/0. Scenario YAMLs for all 7 NY utilities regenerated with seasonal elasticity dicts.

Validation (utils/post/validate_demand_flex_shift.py) — Reproduces the shift analytically and checks: (1) energy conservation, (2) correct direction, (3) match against CAIRO's demand_flex_elasticity_tracker.csv. All 7 NY utilities pass at ε=-0.10. Produces 5 diagnostic plots per utility.

Calibrated results — Full sweep across all 7 NY utilities. Periods YAMLs updated with calibrated values. No-tech summer epsilons: -0.10 (3-hr peak utilities) to -0.14 (5-hr peak utilities). With-tech values are ~1.5-1.7x larger. CenHud calibrated tariffs updated from the ny_20260326_elast_seasonal_tech run.

Docs (context/methods/tou_and_rates/demand_flex_elasticity_calibration.md) — Methodology, full 7-utility results tables for both no-tech and with-tech, corrected Arcturus model description, complete invocation examples.

Reviewer focus

• Arcturus coefficient correction: the with-tech model is a single regression with an interaction term, not a separate model. The intercept is shared (-0.011); only the slope differs (-0.065 no-tech, -0.111 with-tech). The original implementation had a wrong intercept for the with-tech case.
• enabling_tech defaulting to with-tech: an empty enabling_tech cell in the Google Sheet means with-tech (opt-out rather than opt-in). Only explicit false/no/0 selects no-tech. This matches the intended workflow where enabling technology is assumed unless overridden.
• The CAIRO tracker cross-check shows max |Δε| < 0.015. Non-zero differences appear only on the off-peak (receiver) period due to floating-point accumulation order in CAIRO's zero-sum residual. Donor period epsilons match exactly.

Closes #377