Add RCCL clustering playbook for two STX Halo systems

[abdmalik-amd]

• Adds a new playbook for clustering two STX Halo systems using RCCL with vLLM
• Covers VRAM allocation (amd-ttm), vLLM installation with ROCm, Ray cluster setup, and serving Qwen3.5-397B-A17B-GPTQ-Int4 across two nodes via tensor parallelism

Note: This playbook is still being tested. Last validated with RCCL via rocm-systems and vLLM 0.17.0+rocm700 wheels, PR marked as draft accordingly