Remaining issues identified by Konstantinos

[mpompolas]

1. channel coordinates are hardcoded and not absolutely perfect.

We need to go from channel coordinates to NIFTI slice pixel coordinates. This is needed in the case of partial annotations during inference. The coordinates are currently affected by the screen resolution of the computer. Right now the converter uses Brainstorm's functions for creating the 2d topography for each slice.
A more correct approach would be to take the coordinates of the maximum number of channels we can encounter, and assign them uniquely to a 32x32 image (Ivadomed likes images that are multiples of 16. If 32 is not enough, next size is 48x48). Sequentially interpolate on that 32x32 image and create with that process the NIFTI 3D volumes.
NOTE: the interpolation method needs to become a hyperparameter.

2. some model predictions/annotations are overlapping due to the overlapping sliding windows - maybe create a mask before assigning events?

3. Check if loading of the Ivadomed plugin should be done automatically after installation