Replacing PIL and OpenCV

[ShahriarRezghi]

This is the first part of our own image reading functions. This PR adds PNG reading functions using libpng. @fmassa what are your thoughts on these?

• Right now the code is only reading RGB(and not RGBA). Should it stay this way?
• We have two functions. is_png tells us weather the data is a PNG image or not. read_png reads the image and returns a torch::Tensor.
• In case of failure the code returns an empty torch::Tensor.
• How should the tests be? we can write the tests in C++ and compare the read image with OpenCV (or some other library) output (which is not good because it has OpenCV as a dependency). Or we can expose to Python and compare with PIL output.

[codecov-io]

Codecov Report

Merging #1225 into master will decrease coverage by 0.05%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1225      +/-   ##
==========================================
- Coverage   65.64%   65.59%   -0.06%     
==========================================
  Files          74       74              
  Lines        5784     5784              
  Branches      884      884              
==========================================
- Hits         3797     3794       -3     
- Misses       1723     1725       +2     
- Partials      264      265       +1

Impacted Files	Coverage Δ
torchvision/transforms/transforms.py	80.94% <0%> (-0.59%)	⬇️

---

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 8635be9...73c09aa. Read the comment docs.