Add --drop_vector_index to update collection

[jfrancoa]

Summary

Adds weaviate-cli update collection --drop_vector_index <name> to drop the index of a
named vector — the Weaviate 1.39 experimental endpoint
(DELETE /v1/schema/{Class}/vectors/{name}/index), exposed by the Python client in
weaviate/weaviate-python-client#1991 as collection.config.delete_vector_index(). The drop
is destructive and applied asynchronously (a 200 means accepted, not done). The vector can
be re-created afterwards (a fresh, empty index via config.add_vector()) once the drop
finalizes — verified live.

Closes #170

What changed

• update.py / defaults.py — --drop_vector_index <name>, mutually exclusive with
  --vector_index; warns when the server is older than v1.39.0.
• collection_manager.py
  • The drop runs after config.update(). config.update() reads the whole schema and
    PUTs it back, so dropping first would send a vectorIndexType: "none" vector back to the
    server. Ordering is asserted by a unit test.
  • get collection listing fix: a dropped vector's vector_index_config is None; the
    old listing called .vector_index_type() on it and crashed. It now renders none and
    shows every distinct index type across named vectors (e.g. hnsw, none).
  • Re-drop is allowed: re-issuing the drop on an already-none vector returns 200 on the
    server — a no-op while cleanup runs, and a fresh cleanup task if the previous one FAILED
    (the only operator recovery path for a stalled drop). The CLI reports the re-trigger
    instead of blocking it.
• create data auto-skips dropped vectors — after a drop, the server rejects any object
  carrying the dropped vector, so --randomize ingestion failed every batch. Generated
  objects now exclude any named vector whose index is dropped (vectorIndexType: "none"),
  with an informational note (suppressed under --json). Ingestion keeps working; re-ingest
  after re-creating the index.
• Unit tests — drop happy path, --vector_index conflict, unknown/absent vector, listing
  (none / hnsw, none), re-trigger, and the create data skip (one / all / none / --json).
  336 pass.
• Operating skill docs updated (drop semantics incl. re-creation; create data auto-skip).

⚠️ Temporary dependency pin (needs follow-up before release)

requirements-dev.txt and setup.cfg pin weaviate-client to the commit on the
python-client PR #1991 branch
(which adds delete_vector_index() and VectorIndexType.NONE). That PR is not yet merged.
Before a weaviate-cli release this must be reverted to a released weaviate-client>=X. A git
URL cannot be published to PyPI, so this should not merge to a release tag as-is.

Test plan

• make lint
• make test (336 passed)
• Manual, live 3-node 1.39 cluster: drop accepted; sibling search / reads / clean writes
  stay 100% available during cleanup; dropped-vector search rejected immediately; hnsw, none
  shown mid-drop; re-drop of a none vector accepted; --vector_index conflict rejected;
  vector re-created after finalize via config.add_vector() and searchable again; create data skips a dropped vector and ingests successfully (note shown; --json clean).

🤖 Generated with Claude Code

[orca-security-eu]

Orca Security Scan Summary

Status	Check	Issues by priority
Passed	Secrets	0   0   0   0	View in Orca

[orca-security-eu]

Orca Security Scan Summary

Status	Check	Issues by priority
Passed	Infrastructure as Code	0   0   0   0	View in Orca
Passed	SAST	0   0   0   0	View in Orca
Passed	Secrets	0   0   0   0	View in Orca
Passed	Vulnerabilities	0   0   0   0	View in Orca