rename cv::cuda::internal namespace
to cv::cuda::device to prevent conflicts
with cv::internal

* remove overloads with explicit buffer, now BufferPool is used
* added async versions for all reduce functions

cudev is used for new device layer

This reverts commit fba72cb6, reversing
changes made to 02131ffb.

- new reduce implementation (with kepler optimizations)
- saturate_cast via asm command
- video SIMD instructions in element operations
- float arithmetics instead of double
- new deviceSupports function

removed rarely used fixed_size parameter from AutoBuffer type, added optional AutoBuffer* but to cvarrToMat in order to speedup CvSeq->Mat conversion; finished conversion of convex hull and related functions to C++

compilation with CUDA5.0 (#1982, thanks to Rodrigo)
........