[MRG] Preserve weighted low-rank plan shapes under integer indexing

[ktz03]

Types of changes

Bug fix.

Motivation and context / Related issue

Fixes #892.

For get_lowrank_lazytensor(Q, R, d), integer row indexing currently returns (1, m) instead of (m,), and indexing a single entry returns (1,) instead of (). This also affects the lazy plans returned by lowrank_sinkhorn.

Broadcast the diagonal weights directly against the indexed factor, preserving its rank. Parameterized regressions compare shapes and values against the dense product for weighted/unweighted tensors, positive/negative integer indices, mixed indices, columns, blocks, and full matrices. Existing full-plan computations keep the same values.

How has this been tested (if it applies)

• python -m pytest test/test_utils.py test/test_lowrank.py test/test_solvers.py -q: 890 passed, 4 skipped, with NumPy and PyTorch CPU.
• The focused 34-case subset fails 10 shape checks on unchanged production source and passes after the fix; it is included in the 890 tests above.
• Independently checked additional public indexing cases and PyTorch gradients against dense products. Normal indexing controls retain their values and shapes.
• Official pre-commit hooks pass for the changed files.

Windows 11, Python 3.12.14, NumPy 2.5.4, SciPy 1.18.1, PyTorch 2.14.1. Existing transpose/convergence warnings remain. The full repository suite, CUDA, JAX, TensorFlow, documentation build, and performance benchmarks were not run.

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.