Remove device template argument from AtomicArrayExecutionObject
The actual code for AtomicArrayExecutionObject
does not need to be
specialized by the device. The functionality is implemented by calling
the vtkm::Atomic*
methods, which are properly implemented on each
device.