Long-Context Multimodal Understanding No Longer Requires Massive Models: NVIDIA AI Introduces Eagle 2.5, a Generalist Vision-Language Model that Matches GPT-4o on Video Tasks Using Just 8B Parameters