feat: provider-agnostic configuration and timeout changes for init_chat_model

[ansh-info]

• Summary

Issue #474

• Make init_chat_model provider-agnostic by honoring passed args, allowing a supplied chat_model, and avoiding forced ChatOpenAI rebinding.
• Add configurable timeout to the logging wrapper and preserve timeout/factory through tool/structured bindings.
• Keep logging behavior intact while letting custom chat models (e.g., Ollama, OpenRouter, NVIDIA) plug in.

Motivation

• Current implementation ignores init_chat_model arguments, always instantiates ChatOpenAI, and hits a hardcoded 10-minute timeout, breaking local Ollama or other providers.
  Details

• init_chat_model now:

  • Accepts chat_model to plug in any LangChain-compatible chat model.
  • Honors overrides for model, base_url, api_key, temperature, and extra kwargs.
  • Stores a factory so with_config can rebuild without forcing ChatOpenAI.

• LoggingLLM:

  • Adds timeout_seconds (defaults to 10 minutes, configurable).
  • Preserves timeout/factory when binding tools or structured output.
  • with_config rebuilds via the stored factory or delegates to the underlying with_config instead of hard rebinding to OpenAI.