Infrastructure: CRPS and probabilistic scoring framework

## Description
Implement the core probabilistic scoring framework used across Tiers II and III.

## Components
- **CRPS** (Continuous Ranked Probability Score) for ensemble evaluation
- **Spread–skill ratio** for ensemble calibration
- **NRMSE** for deterministic diagnostics
- **Spatial correlation** for pattern evaluation
- Consider **spherical convolutional Wasserstein distance** (Garrett et al., 2024) for future versions

## Notes
- Must handle different ensemble sizes gracefully
- Adapted for larger observational uncertainties in Tier III (proxy data)
- Computational cost (wall-clock time, peak memory) reported but not scored