* avoid using vld1_f16 and vst1_f16 on gcc 4 series (Ubuntu 14.04)
  * guard correctly with #if
  * use static inline

* Common Canny parallelization added. TBB and single thread code removed. Final pass vectorized with SSE2 intrinsics.

* wrong #ifdef replaced with #if

* Merged to actual Canny version

* Merged common parallelized Canny with actual Canny implementation

* Remove 'Mutex *mutex' and pass 'Mutex mutex' from outside to parallelCanny

* Replaced extern Mutex with intern mutable Mutex.

  * stop using 'CUDA' prefix
  * add explanation

Problem is observed with test_loop_times=100
in corner cases like 93x1 or similar

2. fixed parsing of "cat[range_spec]ord[range_spec]" type specification string when using ml::TrainData::loadFromCSV(). Thanks to A. Kaehler for reporting it

  * raise an error when wrong bit depth passed
  * raise an build error when wrong depth is specified for cvtScaleHalf_
  * remove unnecessary safe check in cvtScaleHalf_
  * use intrinsic instead of direct pointer access
  * update the explanation