Bug fix: config.yaml model ensemble syntax not correctly loaded into LLMConfig object

[jvm123]

The new, alternate model syntax introduced in commit f84be60 was not loaded into the LLMConfig object correctly.

Old syntax (still supported):

llm:
  primary_model: "llama-3.3-70b"
  primary_model_weight: 0.8
  secondary_model: "llama-4-scout-17b-16e-instruct"
  secondary_model_weight: 0.2

New n-ensemble syntax:

llm:
  models:
    - name: "llama-3.3-70b"
      weight: 0.8
    - name: "llama-4-scout-17b-16e-instruct"
      weight: 0.2
  evaluator_models:
    - name: "llama-3.3-70b"
      weight: 1.0

The LLMConfig dataclass used default values for primary_model & secondary_model, which the post_init() function gave priority to and overwrote any possible setting in llm:models.

[codelion]

I will merge them over the weekend. Thank you for the fix.