Unified score() dispatcher

[jc-macdonald]

Summary

Provide a single-entry-point dispatcher matching the Python scoring.py pattern, so users can call score(:crps, obs, fct) without importing individual functions.

Proposed API

"""
    score(metric::Symbol, obs, fct; kwargs...) -> Float64

Compute the scoring rule specified by `metric`.

Supported metrics: :crps, :energy, :wis, :brier, :log, :rmse, :mae, :coverage.
"""
function score(metric::Symbol, obs, fct; kwargs...)
    # dispatch to specialized implementations
end

Alternatively, use a type-dispatch approach:

abstract type AbstractScoringRule end
struct CRPS <: AbstractScoringRule end
score(::CRPS, obs, fct) = crps_ensemble(obs, fct)

Design Notes

• Type dispatch is more Julian and extensible.
• Symbol dispatch is simpler for interactive use.
• Consider supporting both: score(:crps, ...) as sugar for score(CRPS(), ...).

Depends On

• Implement CRPS (Continuous Ranked Probability Score) #1 (CRPS), Implement Energy Score (multivariate) #2 (Energy), Implement Weighted Interval Score (WIS) #3 (WIS), Implement Brier Score and Log Score #4 (Brier/Log), Implement point-forecast metrics: RMSE, MAE, coverage #5 (RMSE/MAE/coverage)

Acceptance Criteria

• All implemented metrics accessible via score()
• Informative error for unknown metric symbol
• Docstring lists all supported metrics