BF+ENH: Fixes to DICOM scaling, make frame filtering explicit

Fixes how we handle DICOM scaling, particularly for Philips and
multi-frame files. For Philips data scale factors without defined
units should be avoided, and instead a private tag should be used
to make image intensities comparable across series. For multi-frame
DICOM, it is possible to have different scale factors (potentially
coming from different tags) per-frame. We also prefer scale factors
from a RealWorldValueMapping provided they have defined units.

The base Wrapper class now has a few new attributes and methods to
support this functionality. In particular an attribute `scale_factors`
that provides an array of slope/intercept pairs, and a method
`get_unscaled_data` that will return the reordered/reshaped data but
without the scaling applied. A `vendor` attribute was also added
to better support vendor-specific implementation details.

For the MultiFrameWrapper I also added an attribute `frame_order`
which exposes the order used to sort the frames, and use this
to return the `scale_factors` in sorted order. While implementing
this I kept bumping into issues due to the (implicit) frame filtering
that was happening in the `image_shape` property, so I made this
filtering explicit and configurable and moved it into the class
initialization.

Author: moloney