Fix deprecation warnings

[ChrisRackauckas-Claude]

Summary

• Fix Symbolics.jl deprecation warning: build_function(::Array{<:Equation}...) is deprecated. Use build_function(::AbstractArray...) instead.
• Pass RHS expressions ([eq.rhs for eq in equations(sys)]) instead of Equation objects to build_function in stan_inference.jl
• Update Stan ODE solver integral strings to pass parameters as a vector ([theta_1, theta_2, ...]') instead of individual scalars, matching the new build_function output signature

Details

The deprecated build_function call with Equation objects generated Stan functions with individual real parameters:

vector sho(real t, vector u, real p_1, real p_2) { ... }

The new call with RHS expressions generates Stan functions with a vector parameter:

vector sho(real t, vector u, vector p) { ... }

The Stan ODE solver call (ode_rk45_tol, etc.) is updated to pass a Stan vector [theta_1, theta_2, ...]' to match.

Upstream deprecation notes

• ParameterizedFunctions.@ode_def triggers a @warn from ModelingToolkit about @independent_variables t — this is upstream in ParameterizedFunctions, not fixable here
• The CMDSTAN_HOME warning from StanBase is a test environment issue, not a code issue
• The Turing sample_u0=true test (line 57) is stochastically flaky — it occasionally fails with value ~1e-6 vs expected 1.0, but this is a pre-existing MCMC convergence issue, not related to deprecation

Test plan

• Core tests (DynamicHMC + Turing) pass with JULIA_DEPWARN=error
• Stan string tests pass
• Stan inference tests require CmdStan installation (not available in local env, should pass in CI)

🤖 Generated with Claude Code