Refactor time-integrator storage resolution

[pruliere]

Summary

This PR refactors transient storage resolution so weak forms expose their own storage providers instead of time integrators inferring storage from implementation-specific material attributes.

Main Changes

• Add WeakFormBase.get_storage() and get_dissipation() hooks.
• Update BackwardEuler and GeneralizedAlpha / Newmark to resolve storage through weakform.get_storage().
• Keep first-order weakforms without storage static under BackwardEuler.
• Keep second-order mechanical weakforms explicit: missing density still raises a clear error.
• Move solid inertia resolution into StressEquilibrium.get_storage().
• Add plate/shell storage through section-level density properties:
  • ShellBase.area_density
  • ShellBase.rotary_density
  • homogeneous shell support
  • laminate shell support
• Add beam storage through section-level density properties:
  • BeamProperties.linear_density
  • BeamProperties.rotary_density
• Make GeneralizedAlphaWeakFormSum delegate missing attributes/methods to the wrapped stiffness weakform, preserving APIs such as generalized_strain_operator() after transient wrapping.
• Add focused tests for storage resolution, explicit storage overrides, missing-density errors, plate/shell inertia, laminate density integration, beam section inertia, and first-order no-storage fallback.

Motivation

Previously GeneralizedAlpha fell back to weakform.constitutivelaw.density, which worked for solid StressEquilibrium but was not general enough for structural weakforms. In particular, plate/shell dynamics need area density and rotary inertia, not raw volumetric density.

This PR moves that responsibility to the weakform/constitutive-law layer, where the correct physical storage terms can be defined per formulation.