Dicom importer should modify the imported pixel data if DICOM tags specify this

[jmsmkn]

See comic/grand-challenge.org#1140

[pkcakeout]

Just on technical aspect to this. If we apply these during import, we can actually loose data when the voxel type that we write images as does not support a good enough resolution of the rescaled voxel value space. Example:

Input image type: int8 - contains value 3
Rescale Intercept: 0
Rescale Slope: 0.1 (I _think_ that float is supported here, but not sure - there are other scenarios with different values that we could think off to work purely in the int domain, but they are more complicated)
Target image type: any integer

If we pick an int datatype as target type, then we will loose a bunch of rounded off values that happen during the application of the slope/intercept. Converting everything to float is not a solution either, because some use the integer values to distinguish between classes of data, which does not translate well to floating point values.

I personally, do not like the Rescale-values in DICOM as well because it tends to complicate the interpretation of voxel data quite abit, but they are 100% supported in ITK, and if we pre-apply this, we need to be sure that the consequence, which is potential loss of data, is acceptable to us.

[jmsmkn]

The RescaleSlope and Intercept modifiers have been being applied for the past year (comic/grand-challenge.org#1175). However, if they are applied then we switch to float, otherwise shorts are used covering the data class use case (comic/grand-challenge.org#1244).