docs: bump cascading-retrieval to pinecone 9.1.0

[joerg84]

Why

Continues the docs/ migration off the pre-9.x client. Was pinecone~=8.0.

Changes

• Pin → pinecone==9.1.0.
• 9.x search-hit fix, surgical. Only the reads off actual search Hit objects are renamed to ["id"]/["score"]: print_hits, the dedupe key hit["_id"], and the sort key x["_score"]. The merge_chunks output dicts deliberately keep their "_id"/"_score" keys, so the downstream row["_id"]/row["_score"] reads and the pc.inference.rerank(documents=...) input are unchanged (a blanket rename would have broken those).
• Normalized via ruff + nbstripout.

Verification

All six lint gates pass locally; live test-notebooks validates the runtime.

Follows #598, #599, #600, #601. This completes the cleanly-bumpable docs/ set; the remainder is dependency-blocked (langchain-pinecone <8, pinecone-datasets <4) or requires v3-API rewrites (pinecone==0.2.13).

🤖 Generated with Claude Code

---

Note

Low Risk
Documentation-only notebook changes with no production code or security impact; behavioral risk is limited to example correctness under the new client.

Overview
Updates docs/cascading-retrieval.ipynb for Pinecone 9.1.0 (from pinecone~=8.0), including a pinned install and early Pinecone import in the setup cell.

9.x search hit shape: print_hits and merge_chunks now read id and score on API hits (replacing _id / _score). merge_chunks still emits _id / _score on its output dicts so merge printing and pc.inference.rerank(documents=...) stay unchanged.

Notebook outputs were cleared (execution_count: null, empty outputs) via ruff and nbstripout.

^{Reviewed by Cursor Bugbot for commit cefdaf7. Bugbot is set up for automated code reviews on this repo. Configure here.}