fix: improve randomization and constrain prompts further

fix: implement constrained prompts matching tool-calling behavior

refactor: reduce technical debt by removing unused methods and tool definitions

fix: prioritize extracted metadata over HF dataset fields to prevent mismatches

refactor: proper separation of concerns - aiService as API wrapper, conversationManager builds prompts

fix: make chat responses direct and concise like proper clues

fix: replace tool-calling with simple direct prompts like main branch

fix: prevent AI hallucination and remove problematic tool_choice parameter

fix: correct tool-calling format and contextualization truncation

fix: restore working chat and contextualization logic

fix: restore tool-calling chat prompt engineering

fix: add runtime API key checking and comprehensive debugging

fix: correct AI contextualization method call

fix: restore missing AI methods for word selection and chat

debug: add logging to check API key loading

fix: use complete app.js with full UI logic

fix: resolve all console errors and CSP violations

fix: remove redundant bookDataService initialization

fix: resolve module import errors and CSP issues

feat: add clean aiService.js without hardcoded keys

feat: minimal deployment with essential JavaScript modules and complete CSS