fix sorted ranking so model uses overall sorted retrieved chunks

make executors self-contained, remove redundant pre-filter

[NOTICKET][doc] correct metadata key path in _format_context

[NOTICKET][doc] remove column filter and fallback cap for full-schema approach

[NOTICKET][doc] add sheet-level leg and RRF voting for tabular retrieval

fix: minor returned type if sql writes limit yang melebihi batas

[NOTICKET][db] add sheet-level retrieval and focus LLM schema context to retrieved columns

[NOTICKET] pass orchestrator search_query to sql executor for multi-turn context

[NOTICKET][doc] validate embedding vector for NaN/Infinity in manhattan retriever

fix fallback to fresh retrieval on corrupted Redis cache

[NOTICKET][doc] add guard if filename None

[NOTICKET] now retrieve db tables first, then get column from the obtained tables. reduce k to 5

[NOTICKET][doc] fix aggregate count operation when value_col is not specified

fix: 5 bug fixes on tabular executor

feat: add sheet-level chunk on CSV/XLSX ingestion

fix: increase K in chat endpoint to 10

fix: query executor now use user question as prompt (sebelumnya pakai hasil orchestrator)

[NOTICKET] minor refactoring

[NOTICKET] rrf merge now at router level

fix: fix dedup logic

[KM-533] now also retrieves table level chunk

[KM-533] add table level schema, differentiate with chunk level. expand retrieval result with FK exploration

[KM-455][document] decided methods retrieval for document

[KM-515][document] Make Query for Tabular Type (XLSX & CSV)

[KM-507] now only uses hybrid (cosine and bm25)

[NOTICKET] fix query now use orchestrator msg, rework db pipeline replace ingestion logic

[KM-516][KM-517] add new feature; ai can now see table & column names that have fk relationship with retrieved result

[KM-520] Integrate db query executor pipeline with existing rag retrieve pipeline

[NOTICKET] fix-revert string change

[NOTICKET] db_executor: CTE DML check now walks entire AST root, schema: cast instead of string interpolation

[NOTICKET] fix delete, now can filter by user

[KM-512] connect query executor to user question. add logging for db_executor

[KM-513][document] add convert to parquet if type file is XLSX and CSV

[NOTICKET] rename file name, updated after uv sync

[KM-512] add Pydantic model the LLM fills via function calling in sql_query, and add same signature for db and tabular

[NOTICKET] minor fix in chat.py, add package for query, change schema used to hybrid (cosine+bm25)

[KM-512] create folder for querying from bd/tabular docs

[KM-507] add different methods, now using dense cosine

[NOTICKET] add db_client for querying

[KM-507] add changes to methods

[KM-507] add multiple retrieval method to compare (dense, mmr, bm25, hybrid)

[NOTICKET] fixed multiple sources

[NOTICKET] fix single source to multiple sources

[KM-438][KM-439] framework for knowledge retriever

[NOTICKET][document]: make a clean output to status error unsupported file type

[NO TICKET][document]: delete vector embedding on table langchain_pg_embedding if user delete document on knowledge

[NO TICKET][document]: add updated_at on metadata

[NOTICKET] add updated_at field for metadata & delete old embedding before appending

[NOTICKET] add total token logging

[KM-496] [DED][AI] Fix Response data structure for field sources (db flow)