Upload video and start processing pipeline. Now uses SQLite for storage and Celery for background processing.

New parameters:

• tags: Comma-separated tags for video context (e.g., "product demo,tutorial")
• query: Natural language query for scene matching (e.g., "customer explains features")
• min_duration: Minimum scene duration in seconds (default: 5.0)
• include_captions: Enable captions on exported clips (default: False)
• karaoke_captions: Enable word-by-word karaoke effect (default: False)
• karaoke_color: Hex color for karaoke highlight (e.g., "#cf2a8b", default: brand green "#9DC34F")
• orientation: Target clip aspect ratio for caption positioning: "vertical" (9:16), "horizontal" (16:9), "square" (1:1), or "original" (preserve source orientation)
• processing_mode: Processing mode: "automatic" (detect all scenes), "semantic" (query-based), or "full_video" (no clipping)
• smart_crop_enabled: Enable AI-powered smart crop with subject tracking for intelligent framing (default: False)
• gemini_analysis: Enable Gemini visual scene analysis; when False the pipeline runs Whisper-only and transcript scenes drive clip generation (default: True, still gated by plan entitlement)