Update boundary detection in FSD feature extraction

[FattanePourakpour]

np.argwhere does not trace boundaries accurately; instead, trace_object_boundaries is used.

[cooperlab]

@FattanePourakpour please address the complex curvature problem also mentioned in Issue #1139 and then let's look at getting the tests to pass.

Most tests compare to stored values so we would need to generate new values since the new algorithm is fundamentally changed.

[manthey]

@FattanePourakpour I see you have some commits why say "black". The codebase predates the existence of black and uses flake8 and autopep8. tox -e format will apply the autopep8 rules. They differ from black (for instance, preferring single quotes), and linting will fail with black's rules.

[FattanePourakpour]

@manthey thank you for the clarification. I will do tox -e format and commit again.

[manthey]

@FattanePourakpour The CI is failing because the test_compute_fsd_features no longer matches the stored ground truth file. We can update the ground truth to match the new values if they are what we expect.

[FattanePourakpour]

@manthey Yes, since the new algorithm has fundamentally changed, the ground truth needs to be updated to match the new values for the tests to pass.
Please let me know if there’s anything I can assist with.

[manthey]

@FattanePourakpour I merged master and updated ground truth values. One thing that took me a while was that the trace_object_boundaries expects the edges to be false so that it can avoid checking bounds in the cython code. I added some code to ensure this, as otherwise we were accessing arbitrary memory (and the tests were flaky, though that seems like a success at the detecting it).

If you are content with my changes and have the authority to do so, go ahead and merge.

[cooperlab]

@manthey thank you. We will look at this to see how we can update testing in the future.

This is a fundamental change in the feature extraction so we need to note that the values will be very different in the release (different but correct and likely better).