onemkl linear solver doesn't have a default difference quotient jacobian?

Nobody has claimed this yet.

Hello again,
I finally was able to compile link and execute my model with the onemkl dense linear solver.
The problem now is that at initialization time I get this error
```
[CVLS ERROR] cvLsInitialize
No Jacobian constructor available for SUNMatrix type

[CVODE ERROR] cvInitialSetup
The linear solver's init routine failed.
```

I'm using the following initialization code:

```
/* Create dense SUNMatrix for use in linear solves */
A = SUNMatrix_OneMklDense(y_len, y_len, SUNMEMTYPE_DEVICE, memhelper, &s->qspccQueue, s->sunctx);
if (check_flag((void *)A, "SUNMatrix_OneMklDense", 0)) return s;
s->matrix = A; /* save for deallocation */
// Create the SUNLinearSolver object for use by CVode
LS = SUNLinSol_OneMklDense(s->y, A, s->sunctx);
if (check_flag((void *)LS, "SUNLinSol_OneMklDense", 0)) return s;
s->flag = CVodeSetLinearSolver(s->cvode_mem, LS, A);
if(check_flag(&s->flag, "CVodeSetLinearSolver", 1)) return s;
```

It looks like onemkldense solver doesn't have a default difference-quotient jacobian, like the standard dense/band/lapack solvers.
Is this true? Users are forced to provide a Jacobian through CVodeSetJacFn to use this solver?

As always thank you very much for your help.

Start with the SUNMatrix_OneMklDense and SUNLinSol_OneMklDense entry points, then trace CVodeSetLinearSolver initialization and the CVodeSetJacFn path. Confirm whether this solver provides a default difference-quotient Jacobian like the standard dense, band, and LAPACK solvers; done means the behavior is established and the missing support is addressed if required.

Written by the indexing model from the issue text.