Indicate in QNGOptimizer docs when it does/doesn't work well

[dwierichs]

There are several aspects with respect to which the QNGOptimizer docs could be more thorough/complete.

1. For hybrid cost functions, we currently have a warning in the documentation. Issues that have come up in the past indicate that this should be underlines/refined by providing one or multiple examples of hybrid workflows, and showcasing that there is not necessarily a unique way to run QNGOptimizer for them. We should also include possible solutions, while emphasizing that there is no one-size-fits-all choice.
2. The documentation of the regularization of the metric tensor should be expanded. An example could help to show its purpose (and in particular how QNGOptimizer can have problems if we skip the regularization), but also its effect of watering down the effect of the metric tensor.
3. Overall, the option to provide a custom metric tensor function could be highlighted, maybe through the point(s) above, maybe in addition.