refine densepose coco evaluation code

[Johnqczhang]

1. Refine python code style by adding blanks;
2. Simplify some codes used to compute geodesic point similarity measure (Why write the code so complicated when getting pairwise geodesic distances #202) and evaluation on each image;
3. Rename some variables based on their function/purpose;
4. Add more comments to improve readability.

After refinement, the evaluation results of densepose body UV is the same as previous.

[Johnqczhang]

I'm very sorry for my carelessness in my previous submits which caused very unprofessional operations for a PR. For the convenience, I rollbacked two times to fix all bugs in one submit. The revised version has been fully tested on the "dense_coco_2014_minival" dataset and the same results have been obtained as the provided in MODEL ZOO.

[Johnqczhang]

Closing this PR since it has already been covered in a more comprehensive refinements #215