fix(models): retry async LLM generation

[Ftgn-dpA]

Summary

• apply the existing five-attempt exponential retry policy to async generation in OpenAILLM, LiteLLM, and OpenRouterLLM
• keep async behavior consistent with the corresponding synchronous generation methods
• add offline regression coverage for a transient first-call failure across all three clients

Testing

• uv run --extra dev --extra tools pytest tests/src/models/test_async_retry.py tests/src/models/test_openai_model.py tests/src/models/test_openrouter_model.py -q
• uv run --extra dev ruff check tests/src/models/test_async_retry.py
• uv run --extra dev ruff format --check tests/src/models/test_async_retry.py
• git diff --check

Fixes #275