[Correlation Testing] Half Life Estimation #43

Even if a spread is stationary, it might revert so slowly that it's impractical to trade, i.e. spread with a half-life of 200 days ties up capital for months. There's currently no way to quantify reversion speed.

Implement: A function calculate_half_life(spread) in analysis/spread.py that fits an OLS regression of spread_change ~ spread_lagged (the Ornstein-Uhlenbeck model) and derives the half-life as -log(2) / beta. The result should be added to the metrics dict from calculate_spread_metrics().

Why: Half-life is the primary metric for ranking pairs by tradeability. Short half-life (5-30 days) = actionable. Long half-life (100+ days) = skip. This lets the analysis produce a ranked list of the best pairs rather than treating all 14 equally.