Fix landmark alignment case sensitivity and improve visualization (#84)

[orabe]

This PR resolves issue #84 by adding case-insensitive landmark name normalization and improving landmark visualization in 3D montage plots.

Changes

1. New Function: normalize_landmarks_labels

Location: landmarks.py

Maps alternative landmark names to canonical labels with case-insensitive matching:

"Nz": ["NASION", "Nasion", "nasion", "nas", "Nas", "NAS"],
"Iz": ["INION", "Inion", "inion", "ini", "Ini", "INI"],
"LPA": ["LPA_L", "lpa", "left ear", "Left Ear", "LEFT EAR", "LE", "left", "Left", "L"],
"RPA": ["RPA_R", "rpa", "right ear", "Right Ear", "RIGHT EAR", "RE", "right", "Right", "R"],
"Cz": ["CZ", "cz", "vertex", "Vertex", "VERTEX"]

Handles duplicates by keeping canonical labels and dropping alternatives.

2. Enhanced plot_montage3D Function

Location: montage.py

New Parameter: landmarks: list[str] | None

• None (default): Shows all available canonical registration landmarks (Nz, Iz, LPA, RPA, Cz)
• list[str]: Shows specified landmarks only
• []: Shows no landmarks

Fixes and Improvements:

• Default changed from Nz-only to all 5 registration landmarks
• Fixed PointType enum comparison bug
• Larger landmark markers (s=50)
• Legend positioned outside plot to avoid overlap

3. Testing

Added 12 tests in test_landmarks.py:

• 4 normalization tests (alternatives, duplicates, preservation, empty data)
• 8 plotting smoke tests (modes, defaults, filtering)

All tests pass:

pytest tests/test_landmarks.py -v
# 12 passed in 4.45s

4. Documentation

• Added CHANGELOG entries

Example

from cedalion.geometry.landmarks import normalize_landmarks_labels
from cedalion.vis.anatomy.montage import plot_montage3D

# Normalize landmarks before registration
geo3d = normalize_landmarks_labels(geo3d)

# Visualize with default (all canonical landmarks)
plot_montage3D(rec["amp"], geo3d)

# Or customize
plot_montage3D(rec["amp"], geo3d, landmarks=["Nz", "Cz"])

Fixes #84

[emiddell]

Thank you @orabe!
I changed the logic of normalize_landmarks_labels in case of ambiguities. Instead of dropping duplicates it will raise an error and the user has to sort it out.