Determining unique values is the "abstract" (or categorical)
equivalent of determining the range of interval-valued array.
It only makes sense for arrays with a small number of distinct
values, and this also happens to be when it is computationally
efficient to compute.

This adds methods `GetUniqueComponentValues` and
`UpdateDiscreteValueSet` to `vtkAbstractArray`.
The `GetUniqueComponentValues` method populates a `vtkVariantArray`
you pass with a list of unique values taken on by a component, or,
clears it to an empty array if there are more than
`vtkAbstractArray::MAX_DISCRETE_VALUES` distinct values present.

The unique values are identified by sampling at most log2(n)
tuples for large arrays, so obtaining the summary information --
which is cached once computed -- is an inexpensive operation.
This does mean that rare observations will go undetected.

You may also obtain a list of unique *tuples* instead of unique
component values by passing -1 for the component number.

Unique values are cached in the `vtkInformation` associated with the array.
This commit also adds 2 new key types: one for storing `vtkVariant`
objects as values, and another for storing vectors of variants.
A variant vector is used to store unique tuples and unique component values.

Change-Id: I241cf61372966bba381eb28d14bdc043a54ba197