Add re_max_angle_error to configuration and improve track merging logic

[zitongzhan]

This pull request refactors to the track completion and merging logic in the track_retriangulation.py module. The changes enhance track merging, observation completion, and correspondence handling, as well as configurable options for error thresholds.

Key improvements and refactors include:

Track Merging and Correspondence Handling

• Refactored the merge_tracks function to use view graph correspondences for merging, ensuring only tracks with valid feature correspondences are considered. The merging logic now uses a union-find data structure and updates track ownership for merged observations, increasing merge accuracy and efficiency. [1] [2]
• Added the _build_correspondence_lookup helper function to efficiently map feature correspondences from the view graph, enabling more precise and robust track merging.

Observation Completion and Validation

• Improved the complete_tracks function to avoid duplicating observations, prevent cross-track conflicts, and only add valid observations, using a new _obs_tuple helper for observation identity. [1] [2] [3]
• Replaced the previous reprojection-based observation validation for merging and completion with an angular error check (_point_angular_reprojection_is_valid), controlled by a new re_max_angle_error threshold in the configuration. [1] [2] [3] [4]

Configuration and API Updates

• Added the re_max_angle_error parameter to the COLMAP configuration for finer control over angular error during observation validation.
• Updated function signatures to consistently require the view_graph parameter for functions that depend on feature correspondences, such as merge_tracks and complete_and_merge_tracks. [1] [2] [3]
• Enabled track merging as an optional step, controlled by the enable_merge flag in the triangulator options.

[zitongzhan]

The complete and merge tracks is added, but the retriangulate_underreconstructed_pairs is removed. This need KPI validation