add top-level routing_preferences with selection_policy and model metrics fetch

[adilhafeez]

Closes #848

Summary

• Adds top-level routing_preferences config format (v0.4.0+) where each route groups a pool of candidate models with a selection_policy (cheapest, fastest, random)
• Adds model_metrics_sources — a single HTTP endpoint returning {"cost": {...}, "latency": {...}} for all models, with optional background refresh
• Inline routing_preferences in request body overrides config-defined preferences per-request
• Version enforcement: v0.4.0 forbids per-provider routing_preferences in model_providers; startup validates all referenced models exist

Config example

version: v0.4.0

model_providers:
  - model: openai/gpt-4o
    access_key: $OPENAI_API_KEY
  - model: openai/gpt-4o-mini
    access_key: $OPENAI_API_KEY
    default: true

routing_preferences:
  - name: code understanding
    description: understand and explain existing code snippets
    models:
      - openai/gpt-4o
      - openai/gpt-4o-mini
    selection_policy:
      prefer: cheapest

model_metrics_sources:
  url: https://example.com/model-metrics
  refresh_interval: 300
  # Response: { "cost": {"openai/gpt-4o": 0.005, ...}, "latency": {"openai/gpt-4o": 200, ...} }