feat: reduce thinking budget to lower response time

feat: mitigate citation hallucination

feat: adopt pydantic ai

fix: remove temperature setting

feat: reference citations by a short id

fix(ELN-10): lower temperature for more consistent answers

fix: improve output format instruction following

feat: update gemini to the 05-20 version

feat: add thought summary in the response

fix: history response handling for missing result field

feat: expand source metadata

doc: improve code comments

feat: restful api endpoint