[FIX] Proper Endianness restoration ( [er][et] ShortIntegerArray is broken; here's fixes and tests )

[Yoshiki Ohshima]

  Ned,

  It looks like a nice test.  My tendency towards hacking dominates so
I looked over the ShortIntegerArray...

  For the ShortRunArray case, a ShortRunArray and its 'swapped'
version can behave very differently when the 'value' is zero.  This is
because #size of the ShortRunArray is based on the 'runs' data.

  For example, suppose we have a ShortRunArray instance whose runs and
values look like:

  ((2 0))  "A four-byte instance. the upper 16-bit represent 2 and lower represents 0."

  If I send the #size to this instance, it returns two because the
runs' value is two.  But, if we are loading the serialized version of
this guy from a ReferenceStream, the data looks like at one point:

  ((0 512))

and #size of this guy returns 0.

  I'm saying that for ShortRunArray's case, we might want to add a
non-random version of data to ensure this is properly handled and to
indicate we are aware of this.

  Thank you!

-- Yoshiki