[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 (see: [RFC] Constrained decoding for extension/llm — is it worth doing? pytorch/executorch#19215).

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.

[msluszniak]

Thanks to @jameskanyiri, we have some addional information on tool calling that might be helpful in adding support in the new flow :), see last comment in #429

[jameskanyiri]

@msluszniak @barhanc Happy to help test the implementation once there is a branch available.

[msluszniak]

Linking also this issue to not forget during rewrite: #1314, cc: @lunalinkai