feat: use llm-d well-lit path for generating deployment configurations

[anfredette]

Summary

Replace the current KServe/vLLM Jinja2 templates with llm-d's well-lit path for generating deployment configurations. This aligns the planner with the llm-d ecosystem and ensures generated configurations follow production-tested patterns.

Context

The current deployment configuration generator (src/neuralnav/configuration/generator.py) uses custom Jinja2 templates (src/neuralnav/configuration/templates/) to produce KServe InferenceService YAML. The llm-d project provides a well-lit path for deployment configuration that should be used instead, ensuring compatibility and best practices.

Scope

• Integrate the llm-d well-lit path configuration generation
• Replace or adapt existing Jinja2 templates to use llm-d patterns
• Ensure generated configs are compatible with llm-d deployments
• Update the Configuration Service to use the new generation path

Related Issues

• Add llm-d as deployment target alongside KServe/vLLM #11 (Add llm-d as deployment target alongside KServe/vLLM)
• Add support for configurable routers and gateways #12 (Add support for configurable routers and gateways)