Generate llms files with rocm_docs_generate_llms

[neon60]

Motivation

Enable llms.txt / llms-full.txt generation for the AMD GPU Operator docs, so the site publishes machine-readable documentation following the llms.txt standard (https://llmstxt.org/). This aligns the project with the shared approach used across the Instinct documentation set (k8s-device-plugin, container-toolkit, gpu-cluster-networking, amdgpu-docs, instinct-docs), using the built-in support in rocm-docs-core.

Technical Details

• Enable generation via rocm_docs_generate_llms = True in docs/conf.py, which produces both llms.txt and llms-full.txt from the resolved doctree at build time.
• Add use_download_button: True to html_theme_options for the per-page Markdown download button.
• Bump rocm-docs-core from 1.18.1 to 1.38.0 and add the [llms] extra in docs/sphinx/requirements.in; regenerate requirements.txt (Python 3.10, matching .readthedocs.yaml), which pulls in sphinx-markdown-builder. Existing pins (sphinx-reredirects, Sphinx-Substitution-Extensions) are preserved.

Uses rocm-docs-core 1.38.0 rather than the latest 1.39.0: the published 1.39.0 wheel ships a malformed projects.yaml that breaks the build (already fixed upstream on develop, pending a patch release).

Test Plan

• Regenerated docs/sphinx/requirements.txt with pip-compile under Python 3.10; confirmed the [llms] extra and existing dependencies resolve.
• Ran a full sphinx-build -b html in a clean python:3.10-slim container with the pinned requirements, reproducing the Read the Docs environment.
• Verified llms.txt and llms-full.txt are generated and inspected their content.

Test Result

Build succeeds (build succeeded, 24 warnings) and logs Wrote llms.txt and llms-full.txt. Generated llms.txt (4 KB) has the correct # AMD GPU Operator header and indexes all pages; llms-full.txt (576 KB) has correct per-page sections with prose filtering. The 24 warnings are pre-existing, unrelated to this change.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[spraveenio]

Lgtm