feat: add econometrics research engine v1

[guilhermeFin]

What this adds

A first-class econometrics layer for AlphaForge that complements the existing ML, walk-forward, backtesting, and research-integrity stack.

New research primitives

• OLS with labelled coefficient output
• Classical standard errors
• Newey–West / HAC standard errors with Bartlett weights
• Rolling trailing-window OLS for point-in-time coefficient/regime analysis
• Fama–MacBeth cross-sectional regressions with classical or HAC inference on period premia
• Explicit forward_returns() label construction helper
• Immutable result objects with coefficients, uncertainty, confidence intervals, residuals/fits, sample metadata, and notebook-friendly summary frames

Design constraints

• NumPy/Pandas/SciPy only; no new base dependency
• Does not place trades or weaken existing AlphaForge guards
• Does not duplicate the existing ML ladder (models.py / model_eval.py)
• Does not duplicate walk-forward, purge/embargo, DSR/PSR, transaction-cost, or trial-ledger logic
• Prediction timing remains explicit and auditable: Fama–MacBeth does not silently shift features; researchers construct forward-return labels intentionally

Tests

Adds coverage for:

• recovery of known OLS coefficients
• HAC vs classical inference
• pandas alignment / missing rows
• explicit forward-return timing
• trailing-only rolling regressions
• recovery of synthetic cross-sectional Fama–MacBeth premia
• invalid cross-section configuration

Why

AlphaForge already has a strong research-integrity spine and ML model ladder. This PR fills the main missing piece for a combined Econometrics + AI/ML quantitative research laboratory: interpretable statistical inference that can be compared against nonlinear predictive models under the same downstream honesty framework.

Draft until CI is green and the public API is reviewed.