[MRG] Fix quantized (F)GW solvers

[cedricvincentcuaz]

Types of changes

Fix quantized (F)GW solvers that output mis-ordered indices in final OT plan (Issue #786 ).
Implies modifications of API, providing:

• new parameters part1 and part2 as inputs to quantized_fused_gromov_wasserstein_partitioned. These variables were computed via get_graph_partition but formatted as a list or array-like containing the node assignments. This has been reformatted as a list of array-like, where each array contains the indices of the nodes in the partition, to avoid redundant steps in the previous implementations.

• The construction of the transport plan in quantized_fused_gromov_wasserstein_partitioned is support by two hidden functions _build_full_transport_by_assignment (numpy, torch - mutable) and _build_full_transport_by_concatenation (other non-mutable backends) to leverage a more memory-efficient approach for mutable backends.

• Also fixed a few issues in the example file, where the detailed steps leading to the use of quantized_fused_gromov_wasserstein_partitioned were not leading to the same outputs as the wrapper quantized_fused_gromov_wasserstein because the fallback strategies in the latter for partitioning and representative selection, when networkx and sklearn could be imported, were incorrect.

Motivation and context / Related issue

How has this been tested (if it applies)

• This ordering mismatch was missed by previous tests because marginals were systematically instantiated as uniform ones. I alternated between random marginals and uniform ones to catch potential issues.
• Additional tests were integrated to cover the different situations where NetworkX and/or scikit-learn cannot be imported, with more sensitive fallback strategies in the qFGW wrapper to random partitioning and representative selection schemes -> see test_quantized_gw_only_falls_back_for_missing_dependencies.

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 95.54%. Comparing base (415e95b) to head (7f48113).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #857      +/-   ##
==========================================
- Coverage   96.86%   95.54%   -1.33%     
==========================================
  Files         128      128              
  Lines       26307    26349      +42     
==========================================
- Hits        25483    25174     -309     
- Misses        824     1175     +351     

🚀 New features to boost your workflow:

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