Microphysics architecture: remaining refactoring TODOs

[tapios]

There is significant architectural debt in the microphysics integration. This issue tracks remaining cleanup work, roughly ordered by priority.

• Remove call_cloud_diagnostics_per_stage option: Treat as always true. The per-stage path is the only physically correct one; carrying two code paths adds complexity for no benefit. (The computational expense reasons are no longer a concern.)
• We are supporting the following models:
  • equilibrium + 0M
  • non-equilibrium + 1M
  • non-equilibrium + 2M + P3
    We should remove the separate moisture_model and microphysics_model choices, since they are never separate.
• Consolidate microphysics cache into a single file: Currently, microphysics tendencies and terminal velocities are computed in three separate files:
  • precipitation_precomputed_quantities.jl (grid-mean / quadrature path)
  • diagnostic_edmf_precomputed_quantities.jl (diagnostic EDMF environment)
  • prognostic_edmf_precomputed_quantities.jl (prognostic EDMF environment)
    Move all microphysics cache computations (tendencies + terminal velocities) into a single microphysics_cache.jl, dispatching on turbconv model type. This eliminates cross-file duplication and makes the microphysics boundary explicit.
• Remove duplicate cache allocations for tendency results: ᶜmp_tendency (a NamedTuple field) and the individual tendency fields (ᶜSqₗᵐ, ᶜSqᵢᵐ, ᶜSqᵣᵐ, ᶜSqₛᵐ etc.) store the same data. Use ᶜmp_tendency directly downstream and remove individual fields.
• Remove wrapping of parameter structs such as (cmc,), cmc.activation,) to make them tuples; this won't be necessary with CloudMicrophysics.jl v0.15.5, which makes all those structs broadcastable
• Rename tendency fields for readability: Replace subscript/superscript Unicode names with readable ASCII:
  | Current | Proposed |
  |---------|----------|
  | ᶜSqₗᵐ | ᶜdqlcl_dt |
  | ᶜSqᵢᵐ | ᶜdqicl_dt |
  | ᶜSqᵣᵐ | ᶜdqrai_dt |
  | ᶜSqₛᵐ | ᶜdqsno_dt |
• Harmonize variable names: We use ᶜq_tot_safe for clipped q_tot, but q_liq_rai, q_ice_sno are also clipped.
• Make microphysics tendencies implicit: Add an approximate Jacobian with only diagonal elements for the condensate species (targeting the fastest timescales, e.g., condensation/evaporation). This would improve stability. (Use implicit_microphysics toggle to enable option.)
• Remove remaining two uses of PhaseEquil in utility functions and tests, so Thermodynamics API use is entirely functional.
• Evaluate terminal velocities lazily: Terminal velocities (ᶜwₗ, ᶜwᵢ, ᶜwᵣ, ᶜwₛ) are pointwise functions of local state. Move them out of the cache and evaluate lazily to reduce memory footprint and cache pressure.
• Move toward a functional microphysics API: Reduce reliance on mutating cache writes and if/type-based dispatch in favor of:
  • Pure functions that return results (like apply_1m_tendency_limits)
  • Method dispatch on microphysics scheme types
  • Composable helpers over monolithic set_*! functions
    This reduces code duplication across EDMF variants and simplifies testing.
• Extend SGS quadrature to 2-moment microphysics: Currently, microphysics_tendencies_quadrature_2m only supports GridMeanSGS. Full bivariate quadrature integration (as implemented for 1M) should be extended to 2M schemes.

[trontrytel]

I'm happy to take a stab at the first 4 points. I think t would be good for me to get up to speed with the changed code. Plus, the changes seem easy enough to do. Maybe I can then tag @nefrathenrici for review.

I'm also happy to take on the implicit approximate Jacobian.