AI-assisted submission review, ban system, and steward improvements

[rasca]

Summary

• AI review pipeline: Tier-1 deterministic rules auto-reject clear violations (12.1% of pending), tier-2 Claude Opus analysis handles nuanced reviews via OpenRouter
• Ban system: User ban fields, admin bulk actions, banned user submission blocking, and one-time appeal workflow with steward review endpoints
• Review template improvements: Templates categorized by action type (accept/reject/more_info) with filtered dropdowns in steward UI
• Calibration data: Structured data JSON field on submission notes tracks action, points, staff reply, template ID, and confidence for review quality analysis
• Free-text search: Untagged search terms in steward search bar now search across usernames, notes, and evidence URLs
• Loaddata fix: Signal handlers skip processing during fixture loading to avoid ordering issues

Test plan

• Run python manage.py test contributions.tests.test_review_submissions for AI review pipeline tests
• Run python manage.py test contributions.tests.test_ban_submission_block for ban blocking tests
• Run python manage.py test users.tests.test_ban_system for ban/appeal model tests
• Run python manage.py test contributions.tests.test_calibration_data for calibration data tests
• Verify steward review templates filter correctly by action type in UI
• Test free-text search in steward submissions search bar
• Test ban appeal flow: banned user submits appeal → steward reviews → user unbanned on approval