add frame reducer class and API to compute reduce darks and flats

context

should fix #43 (closed) and #45 (closed)

TODO

• add a FrameReducer class that can be used from the instances of TomoScanBase.
  • handle EDF
  • handle HDF5
• add unit test
  • EDF
  • HDF5
• move from getReducerConstructor to a declarative variable of the class FRAME_REDUCER_CLASS

Question on API:

If no dark and no flat found: should we return None or an empty dict ?

question on processing:

• how to handle output type: should we just add a cast astype() at the end of the processing or should we cast all frames to be used before processing. Error and memory consumption will be very different. For now we only cast it at the end.
• what should be the default dtype ? should we have the same for all or could it be different for EDF (uint16 and int32 to keep compatibility with the existing?) and HDF5 (float32 ?). I think cast should be optional as well.