refactor: clean up LlamaCpp initialization parameters to meet Python syntax standards

fix(rag_core): cast n_ctx to int to resolve pydantic validation error

perf(rag_core): scale context window to 16k and optimize retrieval depth (k=25) for high-fidelity summarization

feat(rag_core): add page-level metadata and fix truncated responses

Update src/rag_core.py

fix(rag_core): expand context window to 8k to accommodate retrieval depth

chore(rag_core): upgrade model to Llama 3.1 8B for improved analytical precision

perf(rag_core): increase retriever k and context window for better PDF analysis

fix(rag_core): add missing PromtTemplate import

fix(rag_core): apply native Llama 3 prompt formatting and add stop tokens to prevent hallucinations

feat(rag_core): enable full CUDA GPU offloading for LLM inference

fix: limit CPU threads to 6 to prevent UI starvation

feat: switch model to Llama 3.2 and establish streaming foundation

feat: Switch local LLM to Phi-3 Mini for better resource management

fix: Correct variable names for model download

feat: Downgrade to Llama 3.2 3B for stability

Engine Swap: Llama 3.1 8B GGUF integrated

Update src/rag_core.py Systempromt LLm answer section

Fix: Groq Import korrigiert, Modell gewechelt

Fix: LangChain Import Pfade korrigiert

feat: rag_core.py zum src-Ordner hinzugefügt