Minor cleanup of roi_align_forward_kernel_impl

[NicolasHug]

We can directly pass n_rois instead of using the intermediate n_threads and output_size names

[datumbox]

I believe that the original author of the method chose to use nthreads to keep the signature the same as roi_align_backward_kernel_impl(). Also note that the same approach is followed at the ps_roi_align_kernel.cpp file.

[NicolasHug]

I think you're right about the original intent, but IMHO the only functions that need to have a common signature are the [ps_]roi_align_forward_kernel, for the dispatcher. The rest can be considered as an implementation detail. Thanks for catching ps_roi_align, I updated that too!

[codecov]

Codecov Report

Merging #3619 (e061359) into master (591c899) will not change coverage.
The diff coverage is n/a.

❗ Current head e061359 differs from pull request most recent head 02ea31d. Consider uploading reports for the commit 02ea31d to get more accurate results

@@           Coverage Diff           @@
##           master    #3619   +/-   ##
=======================================
  Coverage   79.69%   79.69%           
=======================================
  Files         105      105           
  Lines        9824     9824           
  Branches     1583     1583           
=======================================
  Hits         7829     7829           
  Misses       1517     1517           
  Partials      478      478           

Impacted Files	Coverage Δ
torchvision/transforms/functional.py	81.85% <0.00%> (ø)
torchvision/transforms/transforms.py	84.30% <0.00%> (ø)

---

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 591c899...02ea31d. Read the comment docs.

[fmassa]

Fine with me, as we can consider this as an implementation detail.