extend DistributedSampler to support group_size

[stephenyan1231]

summary

For video model evaluation, we sample N clips from a video, and average clip predictions to get a video-level prediction.
Assume, we sample 2 clips per video. The test dataset, which has 4 videos {A,B,C,D} is illustrated below.

[A_0, A_1, B_0, B_1, C_0, C_1, D_0, D_1]

Assume we have 2 gpus. The existing DistributedSampler will distribute clips from the same video to different gpus, and make it difficult to average clip predictions.

GPU 0: 

       [A_0, B_0, C_0, D_0]

GPU 1: 

       [A_1, B_1, C_1, D_1]

We extend ShardDataset to support an optional argument group_size. When group_size=2, which will shard clips below.

GPU 0: 

        [A_0, A_1, B_0, B_1]

GPU 1: 

        [C_0, C_1, D_0, D_1]

This facilitates the averaging of clip predictions.

Unit test

python test/test_datasets_samplers.py

[codecov-io]

Codecov Report

Merging #1512 into master will increase coverage by 0.31%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #1512      +/-   ##
==========================================
+ Coverage   64.34%   64.66%   +0.31%     
==========================================
  Files          83       83              
  Lines        6454     6461       +7     
  Branches      992      992              
==========================================
+ Hits         4153     4178      +25     
+ Misses       2006     1984      -22     
- Partials      295      299       +4

Impacted Files	Coverage Δ
torchvision/datasets/samplers/clip_sampler.py	79.54% <100%> (+23.98%)	⬆️

---

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 937c83a...916801c. Read the comment docs.

[fmassa]

LGTM, thanks a lot for the PR Zhicheng!