Replaced internal variational refinement implementation inside DeepFlow
with a new optimized one, which provides about the same quality while
being 4-6 times faster.

Added variational refinement as a separate class (based on implementation
inside DeepFlow, but significantly accelerated, about 4-6 times faster),
accelerated the main dense inverse search algorithm. Added several new
features including patch mean normalization for increased robustness to
illumination changes and spatial propagation, which often helps to recover
from errors introduced by the coarse-to-fine scheme. Expanded the
documentation, added new accuracy and perf tests. Refactored some of
the already existing optical flow accuracy tests.

Basic interfaces and a partial implementation of the Dense Inverse
Search (DIS) optical flow algorithm without variational refinement. Also
added a python benchmarking script that can evaluate different optical
flow algorithms on the MPI Sintel and Middlebury datasets and build
overall comparative charts.

Update 2

ensure that the double pass through CMakeLists.txt files do not generate double output

see https://github.com/Itseez/opencv/issues/6370

  - remove double-check, make MSVC happy