Fixed bugs conflicting with full non linear variational MHD

[nicolominichelli]

This PR fixes some issues in the variational MHD solver path:

1. Fixed repeated weighted-mass assembly in WeightedMassOperator.assemble() by copying array weights and avoiding in-place multiplication. Previously, spline weights modified persistent geometric weights on every assembly, causing exponential growth of WMMnew and eventual solver breakdown.
2. Assembled the density-weighted velocity mass matrix WMMnew before constructing its diagonal preconditioner in VariationalDensityEvolve.
3. Updated density-dependent mass-matrix preconditioners whenever WMMnew changes, including in the momentum, entropy, and magnetic-field propagators.
4. Added robust happy-breakdown and denominator checks to preconditioned BiCGStab, preventing 0/0 when the intermediate residual has already converged.
5. Fixed the fit_degree index in power_spectrum_2d.