Invalid facet counts returned when adding a filter

[Kerollmops]

Meilisearch returns invalid facet counts in the facet distribution. When a filter is added to a query some of the facet counts increase. This issue is directly related to this Cloud issue (internal link) and you can find the datasets and settings there.

To Reproduce

1. Send a simple query: {"q":"glam","filter":[],"facets":["tags"],"page":0}
2. Look at the facet counts: {"backpack": 9, "bag": 15}
3. Send the same query with a filter: {"q":"glam","filter":[["tags = \"backpack\""]],"facets":["tags"],"page":0}
4. Look at the facet counts increase: {"backpack": 43, "backpacking": 1, "bag": 43}
5. Send the another query with another filter: {"q":"glam","filter":[["tags = \"backpack\"", "tags = \"bag\""]],"facets":["tags"],"page":0}
6. Look at the facet counts increase again: {"backpack": 43, "backpacking": 1, "bag": 67}

Expected behavior
The facet count must be more precise/stable and must even decrease if you constrain the query more with filters. By using page: 0 or a very high limit, the engine must be exhaustive on the facet counts.

Meilisearch version:
v1.8.0

Additional context
On the Cloud, but probably not related.