[Feature] Native LLM-based contextual understanding for media

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request. • Yes The feature Currently, Immich can identify people, objects, locations and other metadata independently. However, it cannot understand the overall context of the scene. Example: video.mp4 Recognised people: • Troy • James Detected objects: • Park • Swing • Slide Today, these are separate metadata points. What I am requesting is something like: Troy and James are playing in the park and discussing why swings are better than slides. Proposed Solution: Introduce an optional LLM-powered media understanding that generates natural language description such as the one mentioned above (for both preferably, but at least photos). It should have • People awareness (eg. Troy is on the swing while James is on the slide) • Overall awareness: The park is next to the hospital. Note: Could b…