Description

This PR restructures and updates the A3 Mega, A3 Ultra, and A4 training support matrices in the main README.md.

Changes

1. Framework formatting: Formatted the Framework column as Framework (Version) (e.g. NeMo (25.07) or Megatron-Bridge (26.02)) to document framework versions clearly.
2. Fixed Broken Links: Corrected folder paths to point to actual directory structures on disk (matching underscores and framework version subfolders).
3. Added Missing Recipes: Documented newer recipes under these platforms that were previously missing:
   • A3 Ultra: Added DeepSeek-V3, GPT OSS 120B, Llama-3-70B, Llama-3-8B, Qwen-3-30B, and Wan-2.1.
   • A4: Added DeepSeek-V3, GPT OSS 120B, Llama-3-8B, Qwen-3-235B, Qwen-3-30B, and Wan-2.1-14B.
4. Cleanup: Removed a duplicate/non-existent Llama-3.1-70B row under A3 Mega.

Verification

• Checked each updated link locally to confirm it correctly resolves to a valid recipe folder in the workspace.