Efficient Fixed Length Vectors and Matrices

The VXL numerics library - vnl - has had its implementations of vnl_matrix_fixed and vnl_vector_fixed upgraded to be fully space efficient.

If you store large numbers of 2D float vectors, you will find a 50% reduction in the storage requirements on most platforms, with no significant loss in performance.

Congratulations to Amitha for doing the work, and the ITK crowd for demanding it be fixed.