AdCare-VLM: Towards a Unified and Pre-aligned Latent Representation for Healthcare Video Understanding