Python binding: add iteration support for Poses and Landmark Observations

[Copilot]

Extends the Python SWIG binding to allow iteration over Poses (already a SharedPtrMap<CameraPose>) and over each Landmark's observations.

Description

Poses already had SPMAP_EXTENSIONS applied, giving it __iter__, keys(), values(), and items() — but this was untested. Landmark.getObservations() returned stl::flat_map<IndexT, Observation> (boost::container::flat_map), which SWIG cannot wrap as an iterable Python object.

Features list

• Iterate over Poses with for pose_id in sfm_data.getPoses(), .keys(), .values(), .items()
• Iterate over Landmark observations via landmark.getObservations(), returning a dict-like Observations object

# Poses
for pose_id, pose in sfm_data.getPoses().items():
    print(pose_id, pose.isLocked())

# Landmark observations
for view_id, obs in landmark.getObservations().items():
    print(view_id, obs.getX(), obs.getY())

Implementation remarks

getObservations() returns stl::flat_map (a boost::container::flat_map alias), which has no SWIG support. Rather than wrapping boost containers, the existing getMapObservations() C++ method (which converts to std::map<IndexT, Observation>) is surfaced via a %extend replacement in Landmark.i. The %template(Observations) std::map<IndexT, Observation> already present in SfMData.i then provides the full dict-like Python interface via SWIG's std_map.i.

Both features are covered by new/updated tests in pyTests/sfmData/.