Guard branching plane computation against zero-norm vectors

[Kalpa08]

Motivation

• Prevent silent propagation of NaNs/infs when the gradient-difference or NAC vectors collapse to zero norm during branching-plane computation.
• Provide a clear, actionable error instead of allowing a division-by-zero to occur deep in the math.
• Improve robustness of the scientific core so downstream plotting/analysis are not fed invalid data.

Description

• Add explicit norm checks in get_branching_plane_vectors and raise a ValueError with a helpful message when either g_ab or h_ab has (near) zero norm.
• Use the computed norms (g_norm and h_norm) when scaling h_ab to g_ab to avoid repeated recomputation and make the intent clearer.
• Add a regression test test_get_branching_plane_vectors_raises_on_zero_norm_inputs in tests/test_real_data.py to lock the behavior for degenerate inputs.

Testing

• Running pytest -q in the default environment initially failed during collection with ModuleNotFoundError: No module named 'conezen' (import-path/install ergonomics issue).
• Attempting python -m pip install -e . failed in this environment due to inability to fetch build dependency setuptools>=61.0 (network/proxy restriction).
• Running PYTHONPATH=src pytest -q (the project test pattern used in CI) succeeded with 4 passed including the new regression test.

---

Codex Task