Miscellaneous fixes and improvements for video reading

[fmassa]

There are a few different fixes in this PR:

• improve should_buffer detection to only happen when divx_packed is set
• fix offset for seek for some files. for some reason they don't seek to the requested position
• add guard for corrupted files
• add guard in VideoClips for videos without stream (fps was None in this case)

The added test uses a file from HMDB51 cartwheel/Turnk_r_Pippi_Michel_cartwheel_f_cm_np2_le_med_6.avi, which is licensed under a Creative Commons Attribution 4.0 International License.

cc @stephenyan1231

[codecov-io]

Codecov Report

Merging #1161 into master will increase coverage by 0.02%.
The diff coverage is 61.11%.

@@            Coverage Diff             @@
##           master    #1161      +/-   ##
==========================================
+ Coverage   65.11%   65.14%   +0.02%     
==========================================
  Files          72       72              
  Lines        5619     5635      +16     
  Branches      867      871       +4     
==========================================
+ Hits         3659     3671      +12     
- Misses       1699     1701       +2     
- Partials      261      263       +2

Impacted Files	Coverage Δ
torchvision/datasets/video_utils.py	83.6% <0%> (-1.4%)	⬇️
torchvision/io/video.py	70.94% <68.75%> (-3.82%)	⬇️
torchvision/transforms/functional.py	71.38% <0%> (+0.28%)	⬆️
torchvision/transforms/transforms.py	81.53% <0%> (+0.98%)	⬆️

---

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 d9830d8...f85379d. Read the comment docs.