[Feature] Decouple face detection/recognition models and add more model choice

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request. • Yes The feature Problem: Currently, Immich bundles face detection and recognition into fixed "packages" (like buffalo_l or antelopev2). This approach is a bit restrictive because it doesn't allow users to optimise for their specific hardware or accuracy needs. Specifically, the default SCRFD detector (while very fast) can struggle with difficult shots. I've noticed many cases where a face is clearly misclassified by the scanner but easily detectable for humans. Proposed Solution Split the machine learning configuration into two independent settings in the UI: • Face Detection Model: To handle the initial pass (e.g., SCRFD, RetinaFace). • Face Recognition Model: To handle the actual embedding/identification (e.g., ResNet50). Additionally, this change would allow adding more models from the [InsightFace Model Zoo]( I s…