Add comprehensive video metadata storage format

[talmo]

Overview

Currently, video lineage metadata (source_video, original_video) is only stored for embedded videos in HDF5 groups. Non-embedded videos have no way to store this metadata, limiting our ability to track video provenance through multiple save/load cycles.

Additionally, the current videos_json format is brittle due to backwards compatibility constraints with SLEAP's cattrs deserialization (see #173).

Proposed Solution

Introduce a new video_metadata dataset that stores comprehensive video metadata in a flexible, versioned format while keeping videos_json unchanged for backwards compatibility.

Design

1. New Dataset Structure

# video_metadata dataset (single JSON string)
{
    "format_version": "1.0",  # Versioning for future extensibility
    "videos": [
        {
            "index": 0,  # Index matching videos_json order
            "filename": "path/to/video.mp4",
            "backend": {...},  # Full backend metadata
            "lineage": {
                "source_video": {
                    "filename": "source.pkg.slp",
                    "backend": {...},
                    "lineage": {...}  # Recursive, with depth limit
                },
                "original_video": {
                    "filename": "original.mp4", 
                    "backend": {...}
                }
            },
            "metadata": {
                # Extensible fields for future use
                "created": "2024-01-01T00:00:00",
                "last_modified": "2024-01-01T00:00:00",
                "user_metadata": {}
            }
        }
    ]
}

2. Key Benefits

• Full Backwards Compatibility: Old SLEAP versions ignore unknown datasets
• Forward Compatibility: Format versioning allows future extensions
• Robustness: Falls back to legacy format if new metadata fails
• Extensibility: Can add timestamps, user metadata, codec info, etc.
• Performance: Single dataset read for all video metadata

3. Implementation Strategy

• Keep writing to videos_json exactly as before
• Write new video_metadata dataset in parallel
• When reading, use video_metadata to enhance videos if available
• Implement depth limits for lineage to prevent infinite recursion

4. Migration Path

1. Read videos_json normally (required)
2. Check for video_metadata and enhance if present (optional)
3. Fall back to checking HDF5 groups for embedded videos (legacy)

Related Issues

• Add option to reference the loaded .pkg.slp file when saving without embedding #188 - Video reference restoration control (current implementation)
• Fix the backend metadata being serialized to SLP for ImageVideo backends #173 - Backend metadata serialization brittleness

Future Extensions

This format enables:

• Storing video FPS, codec information
• User-defined metadata
• Compression statistics for embedded videos
• Edit history/provenance tracking
• Multi-camera synchronization metadata