[WIP] documentation and warnings in (f)gw cg solvers for integer inputs

[cedricvincentcuaz]

Types of changes

• Add note for integer inputs in the documentation of :gromov_wasserstein, gromov_wasserstein2, fused_gromov_wasserstein, fused_gromov_wasserstein2
• Add warnings for integer inputs in : gromov_wasserstein, fused_gromov_wasserstein

Motivation and context / Related issue

• Address issue gromov_barycenters function always returns the zero matrix #530

How has this been tested (if it applies)

• Add tests with integers in test_gromov: test_gromov_integer_warnings, test_fgw_integer_warnings

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

Merging #560 (3d8f4a5) into master (1ece2d8) will increase coverage by 0.01%.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #560      +/-   ##
==========================================
+ Coverage   96.48%   96.49%   +0.01%     
==========================================
  Files          67       67              
  Lines       14773    14816      +43     
==========================================
+ Hits        14254    14297      +43     
  Misses        519      519              

[cedricvincentcuaz]

Hello @rflamary,

For GW (resp. FGW) solvers I set the type of the first variable $C_1$ (resp. $M$) as default type for the outputted transport plan. Do you think that it is enough ? Would it be better to make $T$ inherit the most precise type over all inputs ?
Some use cases: If $p$ and $q$ are not provided there will be instantiated based on this type so the transport plan will inherit it as $G_0 = pq^\top$. If $p$ or $q$ are provided as integers, $G_0$ would have int type so most likely would be a matrix of 0. But providing such marginals really does not make sense. Same argument if $G_0$ is provided by the user with an int type.

[rflamary]

Most precise type of the input is not possible to find easily because of the backend. Its OK to keep the type of M (if aailable) or C with a warning as long as it is clearly stated in thedocumentation.