This PR fixes an inefficiency in Docker AI model downloads where models were always re-downloaded regardless of whether they had changed on the server.

Problem

The common_download_file_single function had special handling for Docker AI models that bypassed the normal HTTP caching mechanism:

// Before: Docker models skipped verification entirely
if (offline || is_docker) {
    return true; // skip verification/downloading
}

// Before: Docker models never sent HEAD requests to check headers
if (!is_docker) {
    // HEAD request logic was skipped for Docker models
}

// Before: Docker models were always downloaded
if (should_download || is_docker) {
    // Forced download even if cached version was current
}

This meant Docker AI models were always re-downloaded on every request, wasting bandwidth and time even when the cached version was still current.

Solution

Removed the special handling that prevented Docker models from using the existing HTTP caching infrastructure:

1. Enabled header checking: Docker models now participate in the same Last-Modified header verification as regular HTTP downloads
2. Preserved authentication: Bearer token authentication continues to work correctly for Docker registry access
3. JSON metadata storage: Last-Modified timestamps are now stored in .json metadata files for Docker models, enabling future cache validation

Benefits

• Reduced bandwidth usage: Docker models are only downloaded when they've actually changed on the server
• Faster model loading: Cached Docker models load immediately without network round-trips for downloads
• Consistent behavior: Docker models now use the same proven caching logic as regular HTTP downloads
• Backward compatibility: No changes to the public API or Docker model specification format

Testing

• Verified the function signature remains compatible with existing Docker model downloading code
• Confirmed JSON metadata structure works correctly for Docker registry URLs
• Validated that CLI tools continue to work with Docker model specifications
• Ensured build process completes successfully with proper C++ formatting

The implementation is minimal and surgical, addressing the exact issue without affecting other functionality.

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.