Changes to vtkDataArray implementation for compatibility with
vtkMappedDataArray. If regular data arrays are used, fast
data iteration via scalar pointers is preserved. If mapped
data arrays are used, vtkTypedDataArrayIterators are used.

Some performance optimizations were made along the way, and
benchmarks added to TestDataArray. The results of the benchmarks
before/after the changes are:

Test                               | Before (s) | After (s) | Speedup
-----------------------------------+------------+-----------+---------
deep copy 10M double --> double:   | 0.0180356  | 0.0179345 |  00.56%
deep copy 10M double --> int:      | 0.0193923  | 0.0108546 |  44.03%
deep copy 10M int --> double:      | 0.0197700  | 0.0125094 |  36.73%
deep copy 10M int --> int:         | 0.0096185  | 0.0088052 |  08.46%
get tuples random access (double): | 0.0111670  | 0.0074788 |  33.03%
get tuples random access (int):    | 0.0106015  | 0.0072431 |  31.68%
insert next tuple (double):        | 0.1429880  | 0.0932984 |  34.75%
insert next tuple (int):           | 0.1445310  | 0.0891607 |  38.31%
insert tuple (double):             | 0.0601416  | 0.0664084 | -10.42%
insert tuple (int):                | 0.0558483  | 0.0630196 | -12.84%
interpolate 2 arrays (double):     | 0.0108466  | 0.0028245 |  73.96%
interpolate 2 arrays (int):        | 0.0114603  | 0.0028706 |  74.95%
interpolate 6 tuples (double):     | 0.0087588  | 0.0055686 |  36.42%
interpolate 6 tuples (int):        | 0.0133066  | 0.0092156 |  30.74%

The insert tuple operation used to just call memcpy, but now it has to
perform some additional type checking, hence the slowdown.

Change-Id: I1d45f2b998d74257a4e1d917e8b01b60fa6ef4e1