Add write_file

[fmassa]

This PR implements a torchscript-compatible way of saving 1d uint8 tensors into a file.

In a follow-up PR, I'll probably remove or refactor write_png / write_jpeg to rely on write_file

[andfoy]

LGTM!

[codecov]

Codecov Report

Merging #2765 into master will increase coverage by 1.74%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #2765      +/-   ##
==========================================
+ Coverage   71.38%   73.13%   +1.74%     
==========================================
  Files          96       96              
  Lines        8311     8319       +8     
  Branches     1293     1293              
==========================================
+ Hits         5933     6084     +151     
+ Misses       1967     1838     -129     
+ Partials      411      397      -14     

Impacted Files	Coverage Δ
torchvision/io/image.py	86.53% <100.00%> (+2.44%)	⬆️
torchvision/io/video.py	69.82% <0.00%> (+1.18%)	⬆️
torchvision/ops/deform_conv.py	70.00% <0.00%> (+3.33%)	⬆️
torchvision/models/detection/rpn.py	93.16% <0.00%> (+4.70%)	⬆️
torchvision/models/detection/roi_heads.py	82.34% <0.00%> (+5.10%)	⬆️
torchvision/datasets/video_utils.py	84.84% <0.00%> (+7.27%)	⬆️
torchvision/ops/boxes.py	93.25% <0.00%> (+7.86%)	⬆️
torchvision/io/_video_opt.py	47.50% <0.00%> (+8.75%)	⬆️
torchvision/__init__.py	78.12% <0.00%> (+12.50%)	⬆️
... and 4 more

---

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 635406c...d1b9431. Read the comment docs.