Multivariate Automatic Differentiation

[Axect]

To calculate exact Hessian via Taylor-mode AD, we should need multivariate formula.

• New structure for Multivariate AD - ADM (temporary name)
• std::ops for ADM
  • Neg, Add, Sub
  • Mul, Div
• PowOps
• ExpLogOps
• TrigOps

[Axect]

Status update: the ADM structure proposed here never landed, and the current AD (Jet<N>) is univariate Taylor-mode. First-order multivariate needs are covered by jacobian() (one seeded pass per column), but exact Hessians still have no path; hessian in numerical/utils.rs is a commented-out stub. Rescoping this issue: the remaining work is a multivariate second-order structure sufficient to compute exact Hessians with cross partials.