Don't assume that we're at start of file at ctor time

BackupDataReader / BackupDataWriter were implicitly assuming that when
instantiated, the underlying fd was positioned at start-of-file.  If one
tried to e.g. open an existing data stream to append further data to it,
things might randomly fail (at read time, possibly when consuming the
stream later) due to incorrect alignment of the data entities: the
appending writer would assume that no padding was needed to achieve
correct alignment, and this might easily be false.

Now the underlying native reader/writer helpers recognize the true
position within the file when constructed, and as a result it's now
safe to e.g. construct a BackupDataOutput for an existing file and
then append to it.

Change-Id: If0484921e687852f923a4b4efabff573a6c16981