Feature request: Rating of matches

Another idea to improve the embedding quality as well as the performance: **Rating of matches**:

- for embedding candidates
- for removal candidates (in case of migration)

Ideas:
- cost function for matches to calculate their cost/quality

Example criteria:
- remove matches from particularly full or particularly empty substrate servers

The cost function (e.g., with input = all attributes of all nodes of a match) could, for example, be learned by ML approaches.

No files, tests, or entry points are named. Start by locating the embedding and migration candidate-selection logic, then determine how match attributes and substrate fullness are represented; done would require an agreed cost or quality function for both candidate types and a validated way to apply it.

Written by the indexing model from the issue text.