The commit "perf improvement convert to using a flat hash" created a
hand-crafted hash table implementation to improve speed.  The commits
"fix UMR in SetAsObjectBase", and "Fixed access beyond bounds bug" later
fixed some memory access problems with the implementation.  However, it
still has at least one memory access violation in rare cases.

Instead of maintaining a hand-crafted hash table implementation it is
simpler to use the existing, well-tested hash_map.  The VTK test case
SpherePuzzleArrows-image creates many pipeline objects that use
vtkInformation instances, so it makes a good test case to compare
performance.  Time spent in vtkInformation::GetAsObjectBase increased by
a factor of about 4 but in absolute terms the time spent in the method
is still small compared to time spent elsewhere in VTK.

Although this change slows vtkInformation a bit, it should avoid memory
access problems altogether.  If performance is a problem in the future
we can then justify spending more time on vtkInformation optimization.