The original was simpler/more readable and we require GCC 5.0 now.

This reverts commit d920a278.

- Replaced Constants.h with std::numeric_limits, see also
  https://jlk.fjfi.cvut.cz/gitlab/mmg/tnl-dev/issues/3
- Made the initialValue method of reduction opeerations static and
  constexpr.

Also added missing 'const' to the getMemoryElement method for host.

Implemented Array:containsOnlyValue.
Restored implementation of ArrayOperations::memoryCompare on host.

Implemented a method Array::checkValue which returns true only if there is an element with given value in the array.

And of course our implementation is faster ;-)

- separate data types for input arrays of binary operations
  (necessary for the equality comparison of two arrays of different
  type)
- added separate template parameter for ResultType, which defaults to
  the RealType
  (the goal is to allow sum of bool vectors into int instead of bool)

- Use math function from TNL/Math.h instead of the standard library
- Many cosmetic changes for readability

The problem is that the __CUDA_ARCH__ macro is defined only in device
code, so it can't be used for configuring the kernel launches.