[MRG] Fix entropic regularization in gcg

[MatDag]

Types of changes

When gcgcalls generic_conditional_gradient, it must be with reg2 = None since the entropic regularization is already taken into account in the inner sinkhorn solver.

Motivation and context / Related issue

This PR solves #758

How has this been tested (if it applies)

n = 100
X = ot.datasets.make_2D_samples_gauss(n, m=[0, 0], sigma=np.eye(2))
Y = ot.datasets.make_2D_samples_gauss(n, m=[1, 1], sigma=np.eye(2))

M = ot.dist(X, Y)
M /= M.max()

a = ot.unif(n)
b = ot.unif(n)

reg = 1e-2

pi1 = ot.optim.gcg(a, b, M, reg, 0., lambda x: 0., lambda x: np.zeros(x.shape), numInnerItermax=1000)
pi2 = ot.sinkhorn(a, b, M, reg, numItermax=1000)

assert np.allclose(pi1, pi2, atol=1e-5)  # now works

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 96.79%. Comparing base (fab3507) to head (f84e1f8).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #817      +/-   ##
==========================================
- Coverage   96.80%   96.79%   -0.02%     
==========================================
  Files         118      118              
  Lines       23721    23721              
==========================================
- Hits        22963    22960       -3     
- Misses        758      761       +3     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[rflamary]

Thanks @MatDag for this PR and catching the problem.

cold you had a very quick test that checks for equality small p small number of iters) so that we will avoid regressions in the future?