docs: clarify RL training GPU requirements by model config

[kibitzing]

• I have personally reviewed this PR and description before asking others to do so. It meets the quality bar I expect from others. I understand that if this PR is perceived as unverified AI-generated code, it will be closed without further explanation.
• I have run tests and confirmed that this code works

---

Description

The previous documentation stated a minimum requirement of 3+ GPUs for GRPO training. However, after #736, the reference model is no longer used by default.

As a result:

• qwen3_1_7b now requires only 2 GPUs for RL training under the default configuration.
• The minimum GPU requirement for RL training has been updated to 2 GPUs.

Additionally, the documentation listed 5 GPUs for Llama 3.1 8B. Based on the current YAML configurations, it requires 4 GPUs (generator + trainer), so the documentation has been corrected accordingly.

This PR updates the relevant documentation to reflect these changes.

Test plan

• Manually verified GPU counts against YAML resource settings:
  • qwen3_1_7b.yaml: generator 1 + trainer 1 = 2 GPUs
  • qwen3_8b.yaml: generator 2 + trainer 2 = 4 GPUs
  • llama3_8b.yaml: generator 2 + trainer 2 = 4 GPUs
• Manually reviewed updated markdown sections for consistency across pages.

[felipemello1]

ty!