Inconsistent fval calculation in gradient check

[m-philipps]

The finite difference calculation in check_grad uses sensi_orders=(0, 1) (or higher) for fval and sensi_orders=(0,) for fval_p and fval_m.

pyPESTO/pypesto/objective/base.py

Lines 563 to 566 in b628397

	# finite differences
	fd_f_ix = (fval_p - fval) / eps
	fd_b_ix = (fval - fval_m) / eps
	fd_c_ix = (fval_p - fval_m) / (2 * eps)

In some cases, e.g. where the gradient calculation via adjoint states requires a smaller step size, this might affect the function value.

To make it consistent, the calculation of fval_p and fval_m should use the same sensi_orders as fval. However, for difficult problems that might introduce some overhead in computation.

[dweindl]

Related to #1487