Recommendations 2.0

Is your feature request related to a problem? Please describe.
The current way to do recommendations (recommended fit) is based on CH. We are seeing that this is often not sufficient and users are often deciding to implement their own Recommendation API.

It's also doesn't support "reverse recommendations", where we do not only recommend A->B but also B->A.
eg. vehicle routing, we currently recommend Vehicles for Visits, but we might want to also recommend Visits for Vehicles.

Describe the solution you'd like
We could implement a more generic approach where we are provided with a plan, a list of potential values (or a filter on the already provided plan) and then something which needs recommending. This could support Recommendations and Reverse Recommendations.

In an initial version, this could just be a Brute Force function, where:

• we loop over all the options
• run score analysis for each of them
• return the best X results.

Most of the time, recommendations are only used with a nearly completed plan anyway, so the list of options is already minimal, but the filter could help to programmatically limit it further. We can look later to improve this.

Describe alternatives you've considered
The current CH approach doesn't do much better with this single assignment.

Start by reading the current CH-based recommendation approach and the score analysis flow mentioned in the issue. Define the scope of the generic and reverse recommendation API, including plan inputs, candidate filtering, brute-force evaluation, and the number of results to return; the issue does not name files or tests.