feat: add asynchronous Omni Moderation for chat inputs and outputs

[GeorgeXie2333]

Summary

• Add asynchronous, OpenAI-compatible Omni Moderation for user prompt text, attached images, assistant text, and generated images.
• Coordinate moderation with active conversation runs, emit live moderation events, and atomically block messages, redact unsafe output, and remove blocked generated files when a policy is triggered.
• Preserve visible moderation alerts across stream completion and conversation reloads.
• Add configurable policies for input/output text and images, including select-all category controls, service probing, concurrency limits, queue capacity, and timeout settings.
• Add admin-only moderation configuration, event inspection, encrypted retained-content review, and anonymous statistics.
• Correct moderation statistics so category breakdown rows do not inflate total hits or produce rates above 100%.
• Add recovery and cleanup paths for delayed verdicts, worker failures, expired retained content, and interrupted streams.

Change type

• Bug fix
• Feature
• Documentation
• Refactor
• Configuration / deployment
• Security hardening
• Other

Affected areas

• Frontend / UI
• Backend / API
• Authentication / authorization
• Conversations / streaming
• Files / RAG / extraction
• Model routing / providers
• MCP / tools
• Billing / payments
• Admin console
• Deployment / Docker / configuration
• Documentation

Verification

• Targeted Go tests for content moderation, conversation streaming, persistence, and secret encryption.
• Go race-detector checks for content moderation concurrency paths.
• Frontend TypeScript typecheck.
• git diff --check.
• Sanitized local streaming API smoke test confirmed a moderation block and an empty blocked assistant result.
• Full repository test and production build suite; not run because validation was limited to targeted checks.

Screenshots, API examples, or logs

Configuration, migration, and compatibility notes

• Adds the content_moderation system-settings namespace with:
  • base_url
  • api_key
  • model
  • timeout_seconds
  • max_concurrency
  • queue_capacity
  • policy_json
  • policy_version
• Defaults to https://api.openai.com/v1 and omni-moderation-latest.
• Moderation remains disabled by default because all policy category lists are initially empty.
• The moderation API key is encrypted at rest and masked in admin responses.
• AutoMigrate creates:
  • content_moderation_events
  • content_moderation_daily_stats
• AutoMigrate adds moderation state, event ID, and category fields to conversation runs and messages. Existing rows use backward-compatible defaults; no data backfill is required.
• Flagged text and isolated images are encrypted at rest. Retained content expires after 30 days, while metadata and anonymous statistics expire after 90 days.
• Configuration and event details require super-admin access. Anonymous aggregate statistics are available to administrators.
• Swagger and the generated TypeScript API contract were updated.
• No new environment variables, public URLs, CORS rules, or Docker changes are required.

Documentation

• Documentation is not needed for this change.
• Documentation was updated.
• Documentation still needs to be updated.

Swagger/API contract documentation includes the new administration endpoints and moderation response fields.

Security and privacy

• No secrets, tokens, credentials, local config, or personal data are included.
• User data access remains scoped by authenticated user context unless an admin-only path explicitly requires broader access.
• Security-sensitive behavior was reviewed, including authentication, authorization, provider routing, file processing, billing, and admin APIs where relevant.

Moderation event details and retained content are restricted to super-admin users. Ordinary administrators receive anonymous aggregate statistics only.

Checklist

• I searched existing issues and pull requests.
• Changes are focused and do not include unrelated refactors.
• Tests or static verification were run where practical.
• User-facing behavior, deployment steps, API contracts, and configuration changes are documented.
• Generated artifacts are included only when this project explicitly requires them.
• Caches, build output, .pyc files, .env files, and local storage data are not committed.