Add method or free function to combine VectorSpaces

[pjb236]

We should provide a method or a free function to allow someone to append modes to an existing vector space. This is needed for optimization, where some methods require combining the forward and adjoint bases to form one basis.

If this is implemented as a method, the call should:

1. take the vector space to append as an argument
2. take in an orthogonalizer, unless we have the vector space own this going forward
3. perhaps include some ability to overwrite the shift vector.

If this is implemented as a free function, the function should:

1. take both vector spaces as inputs, and perhaps modified one in place
2. take in an orthogonalizer, unless we have the vector space own this going forward
3. we might need an argument specifying which shift vector to use for the combined vector space

[pjb236]

Here's a hacky implementation I've gotten to work with existing code. The new version will need to be more eloquent that this in my opinion. Also, it will need to use an orthonormalizer.

forward_basis_tensor = forward_basis.get_basis()
adjoint_basis_tensor = adjoint_basis.get_basis()

forward_basis_matrix = _tensor_to_matrix(forward_basis_tensor)
adjoint_basis_matrix = _tensor_to_matrix(adjoint_basis_tensor)

adjoint_complement = adjoint_basis_matrix - forward_basis_matrix @ forward_basis_matrix.T @ adjoint_basis_matrix
adjoint_modes,_ = np.linalg.qr(adjoint_complement)

basis_matrix = np.concatenate([forward_basis_matrix, adjoint_modes],axis=1)
basis_tensor = _matrix_to_tensor(forward_basis.extents()[0],basis_matrix)

forward_combined_basis = DictionaryVectorSpace(basis_tensor)
forward_shift = forward_combined_basis.get_shift_vector()
forward_shift[:] = forward_basis.get_shift_vector()
adjoint_combined_basis = DictionaryVectorSpace(basis_tensor)