Add landmarks level tests [GSK-2348]

[pierlj]

Added multiples metrics:

• NEs: normalised euclidean distance
• NERFMark: per landmark and per image test whether the prediction is within a given radius around the ground truth
• NERFImagesMean: mean of NERFMark in the image dimension, i.e. per landmark failure rate given the radius average on all images.
• NERFImagesStd: same as above for std.
• NERFMarksMean: mean of NERFMark in the mark dimension, i.e. per image failure rate given the radius average on all marks.
• NERFMarksStd: same as above for std
• NERFImages: proportion of images that have at least failed_mark_ratio landmarks that are not inside the radius around the ground truth.

Added unit tests for these metrics.

Changed the location and name of _calculate_d_outers --> compute_d_outers now in loreal_poc/marks/utils.py.
Also changed a bit the drawing method to display mark with their radius (check ex_8_NERF_metric.ipynb to see an example).

[linear]

GSK-2348 New test on landmarks-level

pre-aggregation of NME: Number of landmarks allowed to not respecting the threshold (radius around the ground-truth)

[rabah-khalek]

Minor changes, otherwise LGTM