feat: Add Scaled CRPS as an accuracy metric

[NateTomasetti]

Continuous Ranked Probability Score (CRPS) is a proper scoring rule to evaluate the prediction of a probability distribution.
It rewards distributions that:

• Have a large amount of density close to the actual value
• Are not spread too wide
• Hits the X% percentile X% of the time

This implements a scaled version of CRPS in the mmm-eval alongside the existing accuracy metrics (MAPE, R2, etc.).
The scaling factor makes this a generalisation of MAPE, so the output value can be consistently compared against a threshold regardless of what kinds of values the data has.

• Regular CRPS is a probabilitstic version of MAE, so the value depends on the magnitude of sales.

---

Note

Introduces scaled CRPS accuracy metric and plumbs posterior predictive distributions from adapters into validation to compute it, with docs, thresholds, tests, and version updated.

• Metrics:
  • Add scaled CRPS computation (crps_one_date, calculate_crps) and include in AccuracyMetricNames/Results and CrossValidationMetricNames/Results (mean_crps).
  • Update thresholds to include CRPS/MEAN_CRPS.
• Adapters:
  • Change BaseAdapter predict/fit_and_predict/fit_and_predict_in_sample to return (posterior_mean, posterior_distribution).
  • Meridian: return posterior mean and flattened distribution; apply holdout mask to both.
  • PyMC: use predict_posterior(combined=True); return mean and distribution (transposed) tuples.
• Validation:
  • Accuracy and CV tests now pass predictive distributions to compute CRPS.
  • Add distribution_to_dataframe utility to reshape distributions for metric calc.
• Docs:
  • Add CRPS to metrics/tests guides, definitions, interpretation, and benchmarks.
• Tests:
  • Update unit/integration/E2E tests for new adapter return types and CRPS calculations.
• Release:
  • Bump version to 0.13.0; update CHANGELOG.md.

^{Written by Cursor Bugbot for commit 9058b5a. This will update automatically on new commits. Configure here.}

[sjmccorm1993]

looks good, could you just

• ensure the unit tests still run and add tests for CRPS specifically
• update metrics.md and tests.md to document the new metric you've added?