Add ResNet implementation with pre-trained weights

[laggui]

Issue #11

This PR adds a Burn implementation for the ResNet family of computer vision models. Implementation and weights are based on torchvision.

TO-DO

• ResNet implementation for resnet-{18, 34, 50, 101, 152}
• Inference example
• Use PyTorchFileRecorder to import weights
• Switch to next Burn release version
  • The Cargo.toml file currently points to a local workspace copy of Burn since parsing the resnet weights with_key_remap requires a fix that came in after the 0.12.1 release and a fix on Candle's side to handle module parameters with pickle
• Automatic download of pre-trained weights (need to re-export download_file_as_bytes for common use in burn)

[laggui]

Addressed the changes for the current version.

Would be great to have a version that includes the fixes but that also depends on Candle releasing a new version with the module parameters fix for pickle.

[nathanielsimard]

LGTM