docs(cli): use hosted VL rerank defaults in CLI example (NVBug 6657245)

[kheiss-uwzoo]

Summary

• Replace the CLI README rerank example that pointed at https://inference-api.nvidia.com/v1/rerank, NGC_INFERENCE_API_KEY, and the retired llama-3.2-nv-rerankqa-1b-v2 model.
• Use the 26.08 hosted vision-language rerank defaults (ai.api.nvidia.com /reranking URL, nvidia/llama-nemotron-rerank-vl-1b-v2, NVIDIA_API_KEY) so customers can run the example with a self-service NVIDIA key.
• Document that --rerank without --reranker-invoke-url uses the local GPU reranker, and that a Cohere-style /v1/rerank URL is a gateway route that needs a gateway-issued credential.

Fixes NVBug 6657245.

Test plan

• Diff is docs-only: nemo_retriever/docs/cli/README.md
• Endpoint and model match nemo_retriever/src/nemo_retriever/operators/rerank.py hosted VL defaults
• QA: ingest a PDF to local LanceDB with hosted embeddings, then run the updated retriever query example with NVIDIA_API_KEY
• Confirm ranking changes versus the no-rerank baseline

pre-draft: leakage, mkdocs --strict, ::a, ::p, ::r on the diff vs main

Base: upstream/main
Files: nemo_retriever/docs/cli/README.md

Check	Result
Leakage (page roles + see [ CTAs)	PASS — no see [ CTAs; CLI README is an allowed documentation path
Allowed paths	PASS — 1 documentation file
mkdocs --strict	PASS for this change — exit 1 from untracked leftover pages (custom-metadata.md, user-defined-stages.md), not this diff
::a audit	PASS — hosted URL/model/key match product defaults; --rerank without invoke URL uses local GPU (verified in query/workflow.py)
::p polish	Applied — customer-runnable hosted example first; gateway /v1/rerank guidance secondary
::r style	95% — no blocking issues

Code drift (not in this docs PR): nemo_retriever/src/nemo_retriever/operators/rerank.py:207 still documents https://inference-api.nvidia.com/v1/rerank in operator help.

PR: this draft