docs(extraction): pass NVIDIA_API_KEY explicitly in Live RAG examples (NVBug 6622594)

[kheiss-uwzoo]

Summary

• Live RAG LiteLLMClient / LLMJudge examples now pass api_key="os.environ/NVIDIA_API_KEY" so the default nvidia_nim model receives the hosted-inference key.
• Published API-key and Python API pages state that LiteLLM does not read NVIDIA_API_KEY for nvidia_nim; omit-api_key looks up NVIDIA_NIM_API_KEY.
• Docs-only. The LiteLLMClient class docstring still lists NVIDIA_API_KEY as auto-read (litellm.py:57-58); that is source drift for a follow-up eng change.

Fixes NVBug 6622594.

Test plan

• Copy the Live RAG README example with only NVIDIA_API_KEY set and confirm the request is sent (not error='transport_error' from a missing LiteLLM key name)
• Confirm api_key="os.environ/NVIDIA_API_KEY" still matches the working OpenAI example earlier on the same README
• Confirm the PR diff is only the three documentation files

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

Base: upstream/main
Files: docs/docs/extraction/api-keys.md, docs/docs/extraction/nemo-retriever-api-reference.md, nemo_retriever/README.md

Check	Result
Leakage (page roles + see [ CTAs)	PASS — no see [ CTAs on these three files; no nimOperator / nvcr.io/nim / installFfmpeg on faq.md, overview.md, or multimodal-extraction.md. Leakage script vs origin/main reported leftover untracked custom-metadata.md (not in this diff; fork origin/main is behind upstream/main).
Allowed paths	PASS — 3 documentation files
mkdocs --strict	PASS for this change — exit 1 from untracked leftover pages custom-metadata.md and user-defined-stages.md, which are not in this diff
::a audit	PASS — 4 claims at ~95%: os.environ/NVIDIA_API_KEY resolves at call time (resolve_environment_reference, LiteLLMClient.complete, test_llm_params.py); default models are nvidia_nim/... (litellm.py:70, judge.py:237); omit-api_key is provider-native (LLMRemoteClientParams); nvidia_nim lookup name is NVIDIA_NIM_API_KEY (evaluation README). Dropped NVIDIA_NIM_ADMIN_KEY (no in-repo evidence).
::p polish	Applied — aligned README wording with api-keys; used "worker" instead of "library"; removed unverified admin-key name
::r style	95% — no blocking issues on the new prose

Code drift (not in this docs PR): nemo_retriever/src/nemo_retriever/models/llm/clients/litellm.py:57-58 still lists NVIDIA_API_KEY as auto-read. Evaluation README LLMJudge nvidia_nim example still omits api_key. transport_error masking is a separate product defect.

PR: this draft

[greptile-apps]

Greptile Summary

The PR updates Live RAG documentation to pass NVIDIA_API_KEY explicitly to LiteLLM-backed NVIDIA NIM models.

• Adds explicit environment-based credentials to the LiteLLMClient and LLMJudge examples.
• Documents the distinction between NVIDIA_API_KEY and LiteLLM’s provider-native fallback variable.
• Cross-references the credential guidance from the Python API documentation.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
docs/docs/extraction/api-keys.md	Adds credential guidance for NVIDIA NIM models and explains explicit versus provider-native environment-variable lookup.
docs/docs/extraction/nemo-retriever-api-reference.md	Directs LiteLLM and LLMJudge users to the environment-reference authentication guidance.
nemo_retriever/README.md	Updates both Live RAG examples to forward the NVIDIA API key through an environment reference.

_{Reviews (3): Last reviewed commit: "Merge branch 'main' into docs/6622594-li..." | Re-trigger Greptile}