Add optional non-negative post-processing for MCNNM baselines

[fedemolina]

Summary

• Add MCNNMPanelSolver.fit(...) dispatch across suggested-rank, fixed-regularizer, and cross-validation paths.
• Add baseline_projection="clip_nonnegative" as optional post-estimation support correction on the final baseline fitted_value + M.
• Preserve raw outputs as canonical (res.baseline, res.tau) and expose projected companion outputs (res.baseline_projected, res.tau_projected, res.projection_diagnostics).
• Add focused tests, a slim local documentation page, and an executable tutorial notebook.

Notes

This is intentionally not framed as a constrained MC-NNM estimator. The projection is transparent post-processing so users can inspect how enforcing a non-negative baseline changes the implied treatment effect.

Validation

• env PYTHONPATH=src poetry run pytest tests/test_mcnnm_baseline_projection.py -q
• env PYTHONPATH=src poetry run pytest src/causaltensor/tests/test_synthetic_class.py::TestSyntheticClass::test_mc src/causaltensor/tests/test_real_class.py::TestRealClass::test_mc -q
• env PYTHONPATH=src poetry run python -m py_compile src/causaltensor/cauest/MCNNM.py tests/test_mcnnm_baseline_projection.py
• Executed tutorials/mcnnm_nonnegative_baselines.ipynb from repo root and from tutorials/ cwd.