Feature Request: Allow Mat and Vec objects to use existing memory

[A-CGray]

Currently, if you pass an existing array to the constructor of an A2D mat/vec, the mat/vec allocates it's own memory and copies the values from the array you supplied. In certain situations it would be useful to have the mat/vec directly use the arrays you've already allocated for the primal value and possibly even the derivatives. This could be useful both for reducing memory usage (e.g in GPU kernels), and in reducing the need to copy out derivative values.

For example:

template<typename T, int N>
void computedfdx(T x[], T dfdx[]){
  ADObj<Vec<T,N>> xVec(x); // At this point we have 4 arrays of length N allocated (x, dfdx, xVec.A, and xVec.Ab)
  ADObj<T> f;
  // Make stack that computes f(x)
  auto stack = MakeStack(...);
  f.bvalue = 1.0;
  stack.reverse();
  // Copy derivative back into dfdx
  for (int ii=0; ii<N;ii++){
    dfdx[ii] = xVec.bvalue[ii];
  }
}

Could instead be implemented as:

template<typename T, int N>
void computedfdx(T x[], T dfdx[]){
  ADObj<Vec<T,N>> xVec(&x, &dfdx); // No additional arrays created here
  ADObj<T> f;
  // Make stack that computes f(x)
  auto stack = MakeStack(...);
  f.bvalue = 1.0;
  stack.reverse();
  // No need to copy derivative back into dfdx
}