Added missing __cuda_callable__ marks

Curiously clang complained about it, but nvcc didn't.

src/TNL/Containers/SharedArray_impl.h

@@ -310,6 +310,7 @@ void SharedArray< Element, Device, Index > :: setValue( const Element& e )
 template< typename Element,
           typename Device,
           typename Index >
+__cuda_callable__ 
 const Element* SharedArray< Element, Device, Index > :: getData() const
 {
    return this->data;
@@ -318,6 +319,7 @@ const Element* SharedArray< Element, Device, Index > :: getData() const
 template< typename Element,
           typename Device,
           typename Index >
+__cuda_callable__ 
 Element* SharedArray< Element, Device, Index > :: getData()
 {
    return this->data;

src/TNL/Containers/StaticArray3D_impl.h

@@ -68,6 +68,7 @@ String StaticArray< 3, Element >::getType()
 }
 
 template< typename Element >
+__cuda_callable__
 inline int StaticArray< 3, Element >::getSize() const
 {
    return size;

src/TNL/Containers/StaticVector_impl.h

@@ -182,6 +182,7 @@ StaticVector< Size, Real >::abs() const
 
 
 template< int Size, typename Real, typename Scalar >
+__cuda_callable__
 StaticVector< Size, Real > operator * ( const Scalar& c, const StaticVector< Size, Real >& u )
 {
    return u * c;