Bounding box ignores inverse flag

[StefanoPierini-Codra]

Problem description

get_data(roi_index) uses single_roi.get_bounding_box() to crop the data. This always returns the bounding box of the ROI shape itself. For an inverse ROI (where the exterior is the region of interest), this crops to the wrong area — the small rectangle inscribing the shape, where most pixels are masked.

The same issue affects compute_geometry_from_obj in sigima/proc/image/base.py, which applies a coordinate offset based on get_bounding_box() — incorrect for inverse ROIs where the data covers the full image.

Proposed solution

In case of inversed ROI, get_bounding_box() must return the whole image.

[StefanoPierini-Codra]

The proposed fix has a limitation: the bounding box sometimes needs to be the circumscribed rectangle to the ROI shape, even with inverse ROIs.

Since the root cause was a naming ambiguity, the "shape's own bounding box" methods were renamed to disambiguate the two concepts going forward:

• get_bounding_box() / get_bounding_box_indices() now unambiguously mean extraction extent (inverse-aware);
• the private helper previously named _get_shape_bounding_box() is now _shape_circumscribed_rect(), and
  ROI2DParam.get_bounding_box_physical() is now get_shape_circumscribed_rect() — both meaning the shape's own geometry, independent of inverse.

Convention: bounding_box = extraction extent, circumscribed_rect = shape geometry.

[StefanoPierini-Codra]

In ROI extraction, in case of inverse ROI, the full image is extracted. To keep it useful, the values masked by the ROI are now imposed to be NaNs. Users can then replace them using Processing -> Level Adjustement -> Replace special values to replace NaNs with a different value.