Update or delete Qdrant point payloads without touching vectors, and reclaim disk after mass deletes

Documented steps

1. Merge/add payload fields while preserving existing ones: POST /collections/{collection_name}/points/payload with {"payload":{"color":"red"},"points":[0,3,10]}, or substitute "filter" for "points".
2. Target a nested payload field with the optional "key" parameter (v1.8.0+).
3. Fully replace a point's payload, dropping unlisted keys: PUT /collections/{collection_name}/points/payload with the same body shape.
4. Remove specific keys: POST /collections/{collection_name}/points/payload/delete with {"keys":["color","price"],"points":[0,3,100]} or a "filter" object.
5. Wipe all payload keys but keep point and vector: POST /collections/{collection_name}/points/payload/clear with {"points":[0,3,100]}.
6. Delete whole points including vectors: POST /collections/{collection_name}/points/delete with {"filter":{...}}.
7. Understand the boundary: set/overwrite/delete/clear payload never touch vector data; only points/delete removes the vector.
8. Deleted points are marked, not purged — they are skipped in queries but still occupy disk.
9. Reclaim space via the vacuum optimizer, configured under storage.optimizers: deleted_threshold (default 0.2, minimum fraction of deleted vectors in a segment) and vacuum_min_vector_number (default 1000, minimum segment size eligible).
10. The optimizer rebuilds affected segments behind a copy-on-write proxy, so the segment stays readable during the rebuild.
11. Reference: https://qdrant.tech/documentation/manage-data/payload/
12. Reference: https://qdrant.tech/documentation/ops-optimization/optimizer/