Removed m_dimension as instance member of TensorStorage with FixedDimensions and instead use the template parameter. This means that the sizeof a pure fixed-size storage is exactly equal to the data it is storing.