- fixed wrong condition (always true)
- replaced unnecessary shift operation: '>> 63' to '< 0'
- used CV_BIG_INT()/CV_BIG_UINT() macros

avoid using of templated 'operator T ()'

* Custom deep learning layers support

* Stack custom deep learning layers

via GCC/Clang __POPCNT__ define

improved performance of v_load_deinterleave(8uC3) & v_store_interleave(8uC3) intrinsics when using SSSE3 instructions.

To dump contents of the last OpenCV error

v_select 'mask' is restricted to these values only: 0 or ~0 (0xff/0xffff/etc)
mask in accuracy test is updated.