Switch to Groq API with llama-3.1-8b-instant model

Use lightweight paraphrase-MiniLM-L3-v2 model + CPU-only torch

Fix embedding model to text-embedding-3-small

Fix langchain deprecation + lazy init RAGExplainer

Use exact versions from working local base env

Add pytesseract, pdf2image, Pillow and system deps

Remove torch/sentence-transformers - using OpenAI embeddings

Initial commit: Resume ATS Score Checker API