Change convention for LSWT dynamical matrix

[kbarros]

With this PR, the dynamical matrix $D$ is rescaled by a factor of 2 so that the quadratic bosonic Hamiltonian can be written,

$$ \mathcal{H} = \frac{1}{2} \Psi^\dagger D \Psi,$$

with $\Psi$ the HP bosons in Nambu spinor form. In this updated convention, the positive eigenvalues of $\tilde{I} D$ are directly the magnon excitation energies.

Apart from aesthetics, there is one real behavioral change in this PR: The effective LSWT regularization strength (with default of 1e-8) is now weakened by a factor of 1/2. That is, the dynamical matrix is now shifted by regularization rather than 2 * regularization. Consequently, the reference data for some "golden" tests must be updated. This modified regularization appears in the final commit of this branch.

[ddahlbom]

To my eye, this all looks good.