Fix Local LLM Quality: Context Stacking & Rope Freq Tweaks

Your local LLMs feel dumb? I fixed local LLM quality by combining a context-stacking prompt technique with specific Ollama `modelfile` parameters. Factual er...