Updated files per bugs 5970369, 5966307, and 5966925

[kheiss-uwzoo]

Summary

This branch brings the 26.03 / NeMo Retriever Library work forward relative to main: library and pipeline behavior, Helm chart and deployment assets, extraction documentation, and supporting API/CI changes. It is a large, multi-topic delta (on the order of 110 files, ~4.8k insertions / ~2.2k deletions vs main).

The most recent commit (NVBug 5970369, 5966307, 5966925) is a focused documentation fix: align audio extraction docs with the Parakeet ASR NIM actually used in docker-compose (nvcr.io/nim/nvidia/parakeet-1-1b-ctc-en-us), replace outdated Riva naming and broken Riva links with NVIDIA Speech NIM documentation, correct support matrix table labels and related links, update quickstart and Python API cross-references, and fix ulimit examples in troubleshooting (remove invalid comma thousands separators in shell snippets).

What changed (high level)

• NeMo Retriever Library (nemo_retriever/): ingest modes, batch/in-process pipelines, PDF and text handling, markdown output, OCR/page elements, rerank support and tests, retriever/recall behavior, benchmarks, and dependency updates.
• Helm: chart and values updates, Nemotron / NIM naming alignment in templates (e.g. page elements, table structure, graphic elements, OCR, embedder, reranker), RTX PRO 4500 overrides, README and gotemplate adjustments.
• Deployment: docker-compose and profile-specific compose files, deployment validation script tweaks.
• Docs (docs/docs/extraction/): extraction guides, support matrix, quickstart, API references, release notes, and related pages refreshed for current product and pipeline defaults.
• API / services: internal NIM interfaces (e.g. OCR, YOLOX), text splitting, Redis ingest service touchpoints.
• Repo hygiene: root README and CONTRIBUTING updates, CI workflow for retriever unit tests, harness/tooling adjustments.