Implementation details and fixes

[brudfors]

• To decrease code repetition and chance of implementation errors some functions/classes can be moved to a utility library:
  https://github.com/csudre/MetricsReloaded/blob/mikael-mods/utils.py
  and imported when needed, i.e.,
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/pairwise_measures.py#L6
  This could also helps in highligting dependencies that would have to be dealt with for a potential MONAI conversion (e.g., skeletonize).
• Parenthesis missing around shape argument, fix:
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/pairwise_measures.py#L428
• Ensure ref and pred are binarised, fix:
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/pairwise_measures.py#L180
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/pairwise_measures.py#L181
• Euclidean distance transform errors as it has two return variables by default, fix:
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/utils.py#L104
• Centre of mass function name same as variable name, fix (has also been moved to utils.py):
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/utils.py#L93
• Maybe include a flag for enabling/disabling the debugging print statements?
• Note that np.percentile interpolates, c.f., Fig 7 in Common... paper that points to a discrete value.
• Using a histogram in all_multi_threshold_values might be cleaner, i.e.
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/prob_pairwise_measures.py#L80
• Added a function x_at_y for less repeated code:
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/utils.py#L107
  i.e.,
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/prob_pairwise_measures.py#L236
• No need to load image data to get nifti shape, i.e.
  https://github.com/csudre/MetricsReloaded/blob/f7f6296468b6556c882cccebb8b6a973d2352e01/mixed_measures_processes.py#L142
• The per_label_dict functions in the classes in mixed_measures_processes.py could perhaps be implemented in a base class, which would lead to less lines of code.

[csudre]

@brudfors - util function is a good idea - will modify to that purpose
Regarding the x_at_y - good point but direction of monotonicity need to be acknowledged
Can't use the histogram due to need to create bins to have equal number of samples