Maintenance of the Mayo clinic human CT dataset loader

[pierocor]

This branch updates the loader and utility code for the Mayo clinic human CT dataset. In particular:

• Update dependency (dicom -> pydicom).
• Add minor fixes due to incongruent metadata and incompatibilities with ODL 1.0.0.
• Implement handling of the flying focal spot via the src_shift_func of the ConeBeamGeometry.

[kohr-h]

Thanks a lot for going though the trouble to keep the Mayo dataset working! Please have a look at my review comments. They should be rather straightforward to address.
Regarding functionality, I trust that the changes you made work for you, so they're kind of tested. After all, it's in contrib.

[kohr-h]

• I suggest to use odl.contrib.datasets.util.get_data_dir() to determine a storage location.
• What are the reasons to change the dataset directory? It would be good to have a small comment on top that says which data it is.
• Please use dir instead of folder. The latter is Windows lingo, and Python uses e.g. is_dir() to check if a Path object represents a directory.

[pierocor]

The Mayo dataset cannot be downloaded directly from a url but it requires its own software. This prevents to use odl.contrib.datasets.util.get_data(filename, subset, url) for local caching. Would this work better?

Suggested change

	# define data folders
	proj_folder = odl.__path__[0] + '/../../data/LDCT-and-Projection-data/' \
	'L004/08-21-2018-10971/1.000000-Full dose projections-24362/'
	rec_folder = odl.__path__[0] + '/../../data/LDCT-and-Projection-data/' \
	'L004/08-21-2018-84608/1.000000-Full dose images-59704/'
	# replace with your local directory
	mayo_dir = ''
	# define projection and reconstruction data directories
	# e.g. for patient L004 full dose CT scan:
	proj_dir = os.path.join(
	mayo_dir, 'L004/08-21-2018-10971/1.000000-Full dose projections-24362/')
	rec_dir = os.path.join(
	mayo_dir, 'L004/08-21-2018-84608/1.000000-Full dose images-59704/')

[kohr-h]

Same question as above: why change the slice?

[pierocor]

It has been adjusted to the example considered (patient L004). In this case, the projection data counts ~32k images and the above slice choice reconstructs a central part of the volume.