[WIP] Faster backend compatible ot.dist

[rflamary]

Types of changes

The current implementation of ot.dist uses the cdist foncton form sipy on non euclidean costs which makes it very slow. This PR is an attempt at making it faster by implementing the most common distances using the backend.

For the moeme,t backend computation are ot always faster

for instanec on a 1024x200 matrix X we have

	cdist (scipy)	dist 0.9.5	ot.dist (numpy)	dist (torch)	dist (tf)
euclidean	0.28965	0.167823	0.107717	0.109713	0.128567
sqeuclidean	0.291672	0.0587267	0.0522944	0.117674	0.0469176
cityblock	0.493404	0.491282	4.61321	1.94563	1.11516
cosine	0.333358	0.333065	0.0471443	0.11068	0.0519573
minkowski	0.294308	0.29352	11.2307	3.10716	11.9075
correlation	0.534073	0.529115	0.0896005	0.123845	0.0723958

Motivation and context / Related issue

How has this been tested (if it applies)

PR checklist

• I have read the CONTRIBUTING document.
• The documentation is up-to-date with the changes I made (check build artifacts).
• All tests passed, and additional code has been covered with new tests.
• I have added the PR and Issue fix to the RELEASES.md file.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.07%. Comparing base (7019709) to head (677c48a).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #701   +/-   ##
=======================================
  Coverage   97.06%   97.07%           
=======================================
  Files          98       98           
  Lines       19955    19985   +30     
=======================================
+ Hits        19370    19400   +30     
  Misses        585      585