[RNE Rewrite] Add LLM chat pipeline implementation

[barhanc]

Implement LLM chat pipeline using the new approach to achieve parity with current implementation. It should support:

• text-only LLMs,
• multimodal LLMs,
• tool calling,
• structured output.

For models like Gemma which require custom runners there should be a separate runner implementation. Also see: #429, #1226, #1225, #1220.

For PoC implementation and discussion see: barhanc/rnet-poc#3.