Enhancement of the STL10 loader

[edouardoyallon]

Closes #167

This PR adds the possibility to use the standard procedure for evaluating a model on STL-10. It simply loads one of the 10 pre-defined folds included in the original tar-file and retains the corresponding data from the training set.

It should be consistent with the previous dataloader, as if no fold is specified, it possibly loads the whole training set (and then leads to an erroneous validation error)

[codecov-io]

Codecov Report

Merging #914 into master will decrease coverage by 0.06%.
The diff coverage is 21.42%.

@@            Coverage Diff             @@
##           master     #914      +/-   ##
==========================================
- Coverage    57.9%   57.83%   -0.07%     
==========================================
  Files          46       46              
  Lines        3696     3709      +13     
  Branches      561      563       +2     
==========================================
+ Hits         2140     2145       +5     
- Misses       1429     1439      +10     
+ Partials      127      125       -2

Impacted Files	Coverage Δ
torchvision/datasets/stl10.py	27.5% <21.42%> (-2.36%)	⬇️
torchvision/datasets/imagenet.py	21.55% <0%> (ø)	⬆️
torchvision/transforms/transforms.py	82.54% <0%> (+0.64%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b384c4e...8231a9a. Read the comment docs.

[fmassa]

Thanks!