Euclidean / SqEuclidean and changes in power

[marcocuturi]

This refers to issue #156.

This PR introduces a few changes:

• oldEuclidean is moved to SqEuclidean and continues wokring with naive **2 and vdot
• Euclidean is now the jnp.linalg.norm(x-y) and is properly differentiated.
• power=1.0 becomes the new default in PointCloud (which is a lot more coherent!).
• Expose cost_fn in PointCloud (I think this should be visible, and useful to debug for instance)

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov-commenter]

Codecov Report

Merging #157 (c863521) into main (b060e2c) will decrease coverage by 5.87%. The diff coverage is 72.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #157      +/-   ##
==========================================
- Coverage   89.84%   83.96%   -5.88%     
==========================================
  Files          51       51              
  Lines        5061     5065       +4     
  Branches      518      519       +1     
==========================================
- Hits         4547     4253     -294     
- Misses        390      660     +270     
- Partials      124      152      +28     

Impacted Files	Coverage Δ
ott/core/potentials.py	82.35% <ø> (-4.71%)	:arrow_down:
ott/tools/k_means.py	92.54% <0.00%> (-3.11%)	:arrow_down:
ott/tools/segment_sinkhorn.py	43.75% <ø> (-56.25%)	:arrow_down:
ott/tools/sinkhorn_divergence.py	61.81% <ø> (-32.73%)	:arrow_down:
ott/geometry/pointcloud.py	68.62% <64.70%> (-15.25%)	:arrow_down:
ott/core/bar_problems.py	74.57% <100.00%> (-9.61%)	:arrow_down:
ott/geometry/costs.py	96.51% <100.00%> (+0.08%)	:arrow_up:
ott/geometry/grid.py	89.65% <100.00%> (-0.87%)	:arrow_down:
ott/core/discrete_barycenter.py	15.58% <0.00%> (-70.13%)	:arrow_down:
... and 28 more