When files get large $DATA is split into several attributes with the
same ID and different runlist_vcn_start. This change supports this by
merging all these attributes into a logical ReaderAt mapped at
different offsets.

Also cleaned up the API and command line:
1. Streams are now referenced in a similar way to TSK (mft_entry-type-id).
2. Now commands can take either a path or mft-type-id notation.
3. Support ADS in path semantics.