chore: sync updated backend to HF

chore: sync updated backend to HF

Fix attack type normalization and safe classification leak in graph

chore: complete ui refactor and security enhancements

Configure 6-class model and deploy to HF Spaces

feat(security): disable OpenAI moderation layer completely per user request

fix(security): add PROMPT_EXTRACTION regex rules to catch explicit system prompt leaks

feat(security): transition firewall to a cumulative risk ensemble architecture

fix(security): enforce pipeline risk threshold over openai moderation boolean

fix(ui): add openai moderation to live monitor breakdown and fix 0 score parsing

feat(security): per-api-key opt-in for openai moderation layer

feat(backend): add optional openai moderation layer

v1.0.0