docs: fix GPT-OSS vLLM install by removing expired PyTorch nightly

[yishangupenn]

Copied from upstream: openai/openai-cookbook#2334
Original author: @sahiee-dev
Originally opened: 2026-01-01

---

Summary

Fix broken GPT-OSS vLLM installation instructions by replacing expired PyTorch nightly dependency with stable CUDA 12.8 wheel.

Motivation

Fixes openai#2330
The Quick Setup section in articles/gpt-oss/run-vllm.md referenced a PyTorch nightly build (torch==2.9.0.dev*) that is no longer available from PyTorch's nightly index. This caused dependency resolution failures when users followed the installation instructions:
No solution found when resolving dependencies: Because there is no version of torch==2.9.0.dev20250804+cu128...

This change:

• Removes the nightly PyTorch dependency
• Installs stable PyTorch 2.9.0 (CUDA 12.8) via direct wheel URL
• Documents Python, CUDA, driver, and GPU requirements upfront
• Adds a rationale note explaining why stable PyTorch is used
  Verification:
• PyTorch wheel URL confirmed available (HTTP 200, 900MB)
• vLLM GPT-OSS wheel index confirmed accessible
• Installation flow validated by community reports in vLLM installation instructions outdated - PyTorch nightly build no longer available openai/openai-cookbook#2330
• Full E2E verification requires CUDA hardware (not available on macOS)

---

For new content

Not applicable — this PR fixes existing documentation only. No new content is being added.