feat: configurable Qwen endpoint + model for Singapore workspace

feat: upgrade synthesis + SQL models to GPT-5.6 Luna Pro

fix: QdrantStore singleton to prevent embedded mode AlreadyLocked error

refactor codebase by removing unused methods and properties from OcrCache, QwenParser, IngestionPipeline, and chat_ui modules. This cleanup enhances maintainability and reduces complexity.

replace Ragas evaluation script with DeepEval implementation for end-to-end evaluation. Update documentation to reflect changes in evaluation strategy and metrics. Remove deprecated Ragas script.

add back the examples in golden.jsonl into 10

implement Phase 6: Observability and Cost Tracking

implement Phase 5: Adaptive Router and SQL Tool

implement Phase 4: Hybrid Retrieval, Reranker, and Grounded Chat

enhance AdaptiveRAG with Qdrant integration and chunking capabilities

add parser enhancements and caching for document conversion

implement Document to Markdown Converter with Gradio UI