[ResNet] Add fine-tuning example

[laggui]

Added a fine-tuning example for multi-label classification on the Planet dataset sample. Inspired by this post when I was looking for multi-label classification datasets examples.

• Refactored resnet-burn to have a virtual workspace and support multiple examples with different dependencies
• Moved imagenet.rs module to the inference example since it is not a strict model requirement
• Added simple fine-tuning example for ResNet-18 with wgpu and tch-gpu feature flags

I consistently get >92% multi-label accuracy when running the example. Training is pretty quick (even on my peasant laptop GPU w/ 6GB VRAM).

Last run:

======================== Learner Summary ========================
Model: ResNet[num_params=11194833]
Total Epochs: 5


| Split | Metric        | Min.     | Epoch    | Max.     | Epoch    |
|-------|---------------|----------|----------|----------|----------|
| Train | Loss          | 0.078    | 5        | 0.381    | 1        |
| Train | Hamming Score | 84.303   | 1        | 97.193   | 5        |
| Valid | Loss          | 0.192    | 5        | 0.412    | 1        |
| Valid | Hamming Score | 86.667   | 1        | 94.078   | 5        |

Training completed in 5m50s
Predicted: ["partly_cloudy", "primary"]
Expected: ["partly_cloudy", "primary"]

[antimora]

LGTM