Implement Dataloader interface

[darsnack]

Implement PyTorch's Dataloader interface.

Highlighted features:

• [x] Support for map-style and iterable-style datasets
• [ ] Custom data loading order
• [ ] Automatic batching
  • [ ] Sampling
• [ ] Multi-process loading
• [ ] Memory pinning (for better GPU support)

Tracking: opus111/FastAI.jl#1

[darsnack]

This issue has been superseded by DataLoaders.jl