fix: llm_results=source_chunks, no history in summarize-all, SQL-based table sources

fix: source attribution via Doc-N mapping + pronoun expansion + person-specificity rule

chore: update app

fix: filter question-only chunks from RAG context to prevent Building_code samples.txt pollution

feat: fix routing (doc wins over generic quantitative terms), add _FINANCIAL_RE, bold formatting in answers, add LangGraph workflow, get_workflow_mermaid

chore: update app

chore: update app

chore: update app

ui: compact document tab action row, all buttons fit one row

deploy: sync and update

Initial commit for HF Space