Time-only universal Helmholtz classification (work package C)

{Tests (wpC/time_only_specialization_tests.txt).} The intertwining is tested exactly on $400$ random rational polynomials (seed $20261006$) in all four directions, and for random iterated multi-indices $|\beta |\le 4$. It is tested by an independent sympy route with $\sigma =\sigma (x^0)$, $\chi (x)$ and $h(x)$ as functions, and on $960$ actual collapse and formal adjoint-defect expressions ($120$ contact variables $\times$$4$ directions). A negative control, a ${D_1}^{(t)}$ that differentiates $\sigma$, violates the identity. Route 1 versus route 2. Route 1 is $S_t\circ {\mathrm{Collapse}}_{\mathrm{formal}}$, and $S_t$ applied to the formal adjoint defect. Route 2 applies $S_t$ to the graph table and then builds the collapse by iterated single ${\partial }_w$ (Leibniz on ${\chi }_{\mathrm{I}}$ and $h$ only) and the adjoint by iterated $D^{(t)}$ on (coefficient $\times$ test field), with no binomial closed forms. The two routes agree exactly over $\mathbb{Q}$ on all $3321$ columns of both maps: $446292$ nonzero entries for ${\mathrm{Collapse}}_t$ and $573582$ for $H_t$. In the $\chi$-plateau variant they also agree, with $91568$ and $130110$ entries.