Add angle spline interpolation postprocessing pass

[BastianHavers]

Summary

This PR adds a new opt-in postprocessing pass AngleSplineInterpolationPass that derives bounding box heading angles by fitting a cubic spline to each vehicle's trajectory.

Motivation

Existing rotation adjustment relies on frame-to-frame movement vectors, which can be noisy. Fitting a spline to the entire trajectory provides a globally smooth estimate of heading direction.

Changes

New feature: --angle-spline-interpolation S

Invoke from the CLI with --housekeeping --angle-spline-interpolation <S> where S is the splprep smoothing factor (s=0 = exact interpolation, larger = smoother curve).

Algorithm

1. Extract (x, y) bounding box centre coordinates per tracked object
2. Remove consecutive duplicate positions (vehicle stopped) to avoid degenerate splines
3. Fit a parametric cubic spline via scipy.interpolate.splprep
4. Obtain tangent vectors via scipy.interpolate.splev(der=1)
5. Set bbox angle to atan2(dy, dx) of the tangent
6. Forward-fill angles to stationary frames

Objects with fewer than 4 unique positions are skipped gracefully.

Pipeline placement

Runs after RotationAdjustmentPass (step 8b) and before DuplicateRemovalPass, so IoU-based duplicate detection uses the corrected angles.

Provenance

The flag and smoothing value are recorded in build_arguments_string() and captured by the dataprov provenance chain.

Files changed

• pyproject.toml — added scipy>=1.10.0 to markit dependency group
• markit/markitlib/postprocessing/_passes/angle_spline_interpolation.py — new pass
• markit/markitlib/postprocessing/_passes/__init__.py, passes.py, __init__.py — exports
• markit/run_markit.py — CLI arg, pipeline wiring, provenance tracking
• markit/markitlib/config.py — config attribute
• markit/tests/test_angle_spline_interpolation.py — 16 unit tests

Tests

All 138 existing tests continue to pass. 16 new tests cover: straight-line, curved trajectory, stationary frame forward-fill, too-few-points skip, width/height normalisation, housekeeping tag, statistics, smoothing factor effect, multiple objects, and the deduplication helper.