DlibOptimizer -- don't hardcode solver_epsilon

[dweindl]

DlibOptimizer currently has a hardcoded solver_epsilon (the magic 0.002):

pyPESTO/pypesto/optimize/optimizer.py

Lines 672 to 678 in ead29b3

	dlib.find_min_global(
	get_fval_vararg,
	list(lb),
	list(ub),
	int(self.options["maxiter"]),
	0.002,
	)

My current interpretation of that parameter (documentation is a bit sparse) is that it's the minimum absolute objective function improvement (https://github.com/davisking/dlib/blob/6d29e0c7d41dfdffb8c0a0f83995caae84df5083/dlib/global_optimization/global_function_search.cpp#L754). For that, this value is potentially rather high. I am not sure if we should override dlib's default at all (their default is 0), but at least it should be configurable by the user.

[PaulJonasJost]

Agreed, will make it something the user can change but with default of 0