TAP-CT: 3D Task-Agnostic Pretraining of Computed Tomography Foundation Models