refactor: instructor_llm_factory merge with llm_factory

[anistark]

Consolidates dual LLM factories into a single, explicit llm_factory() API. Removes instructor_llm_factory() entirely.

Changes

Core Changes

• Unified llm_factory(): Single factory supporting multiple LLM providers (OpenAI, Anthropic, Cohere, Google, LiteLLM)

  • Requires explicit parameters: model, provider, client
  • Returns: InstructorBaseRagasLLM (structured outputs only)
  • Clear error messages for missing parameters

• Removed instructor_llm_factory(): Deleted entirely (~97 lines)

  • Functionality merged into unified llm_factory()
  • Removed from public exports

API Change

Before:

from ragas.llms import instructor_llm_factory
llm = instructor_llm_factory("openai", model="gpt-4o", client=client)

After:

from ragas.llms import llm_factory
llm = llm_factory("gpt-4o", client=client)  # provider defaults to "openai"

Migration for Users

# Old code
llm = llm_factory("gpt-4o-mini")

# Error message:
# ValueError: llm_factory() requires a client instance.
# Text-only mode has been removed.

# To migrate:
from openai import OpenAI
client = OpenAI(api_key='...')
llm = llm_factory('gpt-4o-mini', client=client)

Notes

This is a breaking change for the public API, but justified because:

1. Old text-only factory was already deprecated
2. All new metrics require structured outputs (InstructorLLM)
3. Unified API is simpler and more maintainable
4. Clear error messages guide easy migration