Bootstrap Your Own Views: Masked Ego-Exo Modeling for Fine-grained View-invariant Video Representations