[Platform] Add qualified runtime dispatch to Edge-LLM

Summary

Allow Model Connect to dispatch supported workloads to Edge-LLM when a qualified Edge-LLM implementation is available, while retaining the existing Model Connect implementation as the fallback.

Scope

• Discover Edge-LLM runtime capabilities.
• Match qualified profiles using model, revision, platform, precision, layout, and runtime versions.
• Adapt artifacts and data layouts when required.
• Translate configuration and inference requests.
• Normalize outputs, errors, and metrics.
• Fall back cleanly when no qualified Edge-LLM profile exists.
• Make the selected runtime visible in logs and trtmc inspect.
• Add unit and end-to-end tests.

Acceptance criteria

A supported model dispatches to Edge-LLM only when its complete profile matches. Unsupported configurations continue through the native Model Connect path without user-visible breakage.

Start at the trtmc inspect entry point and trace the existing native Model Connect dispatch path. Identify how runtime capabilities, qualified profiles, artifacts, requests, outputs, errors, and metrics are represented before scoping the Edge-LLM integration. Done means complete profile matches use Edge-LLM, unsupported configurations retain the native path, and unit and end-to-end tests cover both cases.