docs(cli): use hosted VL embed model in hosted NIM examples (NVBug 6657245)

[kheiss-uwzoo]

Summary

• Follow-up to docs(cli): use hosted VL rerank defaults in CLI example (NVBug 6657245) #2576. The rerank URL/model/key fix is still correct.
• nvidia/llama-nemotron-embed-1b-v2 now returns HTTP 410 on integrate.api.nvidia.com (EOL 2026-08-25), so the hosted ingest example fails before the rerank example can run.
• Change --embed-model-name to nvidia/llama-nemotron-embed-vl-1b-v2 in the hosted ingest and rerank examples only. Leave local Hugging Face checkpoint mentions unchanged.

Fixes the remaining runnable-example gap on NVBug 6657245.

Test plan

• Diff is docs-only: nemo_retriever/docs/cli/README.md (2 token replacements)
• Hosted examples match the support-matrix default NVCF embed model nvidia/llama-nemotron-embed-vl-1b-v2
• Fast-text and checkpoint-list mentions of llama-nemotron-embed-1b-v2 are unchanged
• QA: copy the hosted ingest then rerank examples verbatim on a CPU-only host with NVIDIA_API_KEY

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

Base: upstream/main (afbddc61)
Files: nemo_retriever/docs/cli/README.md

Check	Result
Leakage (page roles + see [ CTAs)	PASS — no CTA/link changes
Allowed paths	PASS — 1 documentation file
mkdocs --strict	not re-run this follow-up; snippet include is unchanged except two model IDs
::a audit	PASS — VL embed ID matches VL_EMBED_MODEL and the Default NVCF endpoints table
::p polish	none needed — one-token model ID swap
::r style	100% — identifiers only

Code drift (not in this docs PR): operators/rerank.py:207 still names inference-api.nvidia.com in a docstring. QA marked that optional and not a closure blocker.

PR: this draft (follow-up to #2576)