feat: add production prompt engineering patterns with benchmarks

[rehan243]

Summary

Adds a collection of production-tested prompt patterns with:

• Chain-of-thought, few-shot CoT, structured extraction, RAG grounding
• Expected improvement benchmarks for each pattern
• Simple render API for template usage

Motivation

The guide covers theory well but could benefit from more executable production patterns with measured improvements.

Based on patterns from deploying LLM features at enterprise scale.

[vercel]

@rehan243 is attempting to deploy a commit to the DAIR-AI Team on Vercel.

A member of the Team first needs to authorize it.