v0.3.0

All new sharding design, making sharding optional and decoupled from dataloading.

Updated examples to be more feature-complete.

Removed redundant datasets and moved towards a trait-system to make datasets easier to customize.