docs: add inference benchmarks and model capabilities

[arpannookala-12]

Summary

• Add Inference Benchmarks section with performance comparison across Ollama, vLLM, Intel OPEA EI, and OpenAI (gpt-4o-mini)
• Add Model Capabilities section with detailed specs for Qwen3-4B-Instruct-2507 and GPT-4o-mini, plus a side-by-side comparison table
• Link Intel OPEA Enterprise Inference repo in benchmarks
• Update TOC with new sections

Test plan

• Verify markdown tables render correctly on GitHub
• Verify TOC links resolve to correct sections
• Verify Intel OPEA EI link works