fix(discretization): fold symbolic initial conditions for MTK v1.45 compatibility

[utkuyilmaz1903]

Root Cause

MTKBase v1.45.1-era varmap_to_vars now strictly enforces SU.isconst checks on u0 map values. Meanwhile, MOL's generate_ic_defaults builds u0 values via substitution (here via pde_substitute), which leaves function-call ICs as unfolded symbolic trees rather than literal constants (e.g., cos(0.108) remains a symbolic tree). This mismatch triggers Initial condition underdefined errors during discretize(). ICs containing function calls fail, while plain/constant ICs (e.g., Burgers, Integrals) continue to pass.

Fix

• generate_ic_defaults.jl: Wrapped substituted u0 values in Symbolics.symbolic_to_float. This safely folds constant expressions into numbers while preserving genuinely symbolic/parametric expressions for MTK to resolve. Also removed a dead out = ... block.
• MethodOfLines.jl: Imported symbolic_to_float.
• qa.jl: Added :symbolic_to_float to the ExplicitImports ignore list.

Verification: Verified locally on failing groups. Diffusion, Components, Convection_WENO, Sol_Interface, Complex, and Wave_Eq_Staggered are now green. No regressions in previously passing groups (Burgers, Integrals, Stationary).

Reviewer Note

• QA / Non-Public API: symbolic_to_float is not public API, so I added it to the QA ignore list (following the :diff2term precedent). Since public APIs like simplify or substitute no longer fold constants in Symbolics v7, this was the safest workaround. If there is a preferred public alternative, please let me know and I will gladly switch! (Note: I did not bisect the exact MTKBase commit, but the failure window aligns perfectly with the v1.45.1 release and the new SU.isconst checks).
• DAE Group: Still fails, but due to an independent cause. OrdinaryDiffEq's stricter CheckInit now rejects the O(dx²) residual in MOL_1D_PDAE.jl. Passing initializealg = BrownFullBasicInit() fixes it locally (matching the schroedinger.jl precedent), but I left it out of this PR to strictly contain scope.