Fix sample questions, remove upload feature, fix README accuracy

Remove upload feature: broken with scanned PDFs, replaced by fixed corpus

Add RAGAS eval results: 27-question golden set, avg 0.811

Add checkpoint resume + batch upsert to handle quota limits

docs: chuyển demo links lên đầu, làm rõ data offline và eval command

ui: thêm latency counter và sửa empty state text

fix: sửa docker port, bổ sung gitignore, thêm Hạn chế & Hướng phát triển vào README

docs: thêm màu vào Mermaid pipeline diagrams

docs: chuyển pipeline diagrams sang Mermaid

fix: eval dùng 8b-instant + checkpoint resume + cắt context để tiết kiệm token

fix: use llama-3.1-8b-instant for eval to avoid shared TPD exhaustion

fix: query rewriter falls back to original query on rate limit error

data: expand golden set từ 12 lên 30 câu hỏi

ui: show ColPali unavailable message on HF Spaces

feat: show rewritten query in chat UI

feat: expand source to show retrieved text chunk

fix: graceful fallback when colpali_engine not installed

fix: rebuild BM25 index from Qdrant if pkl not found

fix: download bm25_index.pkl from HF repo at build time

fix: remove bm25_index.pkl COPY from Dockerfile (uploaded via HF API)

feat: add compare pipeline tab + has_context badge in UI

docs: add key results to intro, add GitHub link

chore: track pkl files with git lfs

ui: update sample questions to financial report queries

feat: contextual retrieval eval + update README results table

feat: contextual retrieval with llama-3.1-8b-instant, Recall 0.40→0.72

docs: chuyển README sang tiếng Việt

feat: add ColPali eval script + update comparison results in README

docs: update README with ColPali comparison, eval results, architecture

feat: add ColPali visual retrieval + contextual retrieval + PDF eval

feat: dark mode, mobile responsive, copy button, skeleton loader, sample questions

docs: add live demo link and architecture decisions

fix: add nginx buffer flush padding for SSE on HF Spaces

fix: async SSE generator with blinking cursor for streaming

feat: streaming response via SSE with token-by-token display

fix: add python-multipart for file upload

feat: PDF upload with drag & drop, ephemeral in-memory search

perf: preload reranker model at startup

fix: remove *.pkl from dockerignore

fix: explicit bm25 copy path to bust cache

fix: allow bm25_index.pkl in Docker build

fix: copy bm25_index.pkl into Docker image

fix: commit bm25_index.pkl for HF Spaces deployment

debug: print QDRANT_URL

remove debug log

debug: print GROQ key prefix in logs

fix: strip whitespace from API keys

fix: remove local hf_models copy, let model download on startup

fix: add HuggingFace Spaces configuration header

docs: replace roadmap with Future Work section