OmniActions: Predicting Digital Actions in Response to Real-World Multimodal Sensory Inputs with LLMs