docs(host/quickstart): add MiniMax M2.7 and reference repo node-configs

[tcharchian]

Summary

Follow-up to gonka-ai/gonka#1260 (MLNode 3.0.14 & MiniMax M2.7). Adds MiniMax M2.7 as the 3rd governance-approved model in docs/host/quickstart.md and surfaces the ready-made node-config-*.json files that ship in gonka/deploy/join/.

Changes

• Supported models table — add MiniMaxAI/MiniMax-M2.7 as the 3rd approved model.
• Proposed Hardware Configuration table — add MiniMax row (4×A100 / 4×H100 / 2×H200 / 2×B200, ~320 GB VRAM per ML Node).
• New "Reference deploy configs in the repo" tip — lists every deploy/join/node-config-*.json shipped for Qwen, Kimi, and MiniMax, so hosts can copy a ready file instead of writing one from scratch.
• Edit Inference Node Description section — new tabs, each mirroring the corresponding JSON in gonka/deploy/join/:
  • Qwen — 8×B200
  • Kimi — 8×H200 (FLASHMLA, tp=8)
  • MiniMax — 4×A100 (marlin MoE + note about VLLM_USE_FLASHINFER_MOE_FP8=0)
  • MiniMax — 4×H100 (FLASHINFER + fp8 kv-cache)
  • MiniMax — 2×H200 (FLASHINFER + fp8 kv-cache; matches the configuration used to record MiniMax PoC golden vectors)
  • MiniMax — 2×B200 (FLASHINFER_TRTLLM MoE + fp8 kv-cache)
• Pre-download Model Weights — new MiniMax M2.7 tab with huggingface-cli download and a note that MiniMax requires MLNode 3.0.14+ (image ghcr.io/gonka-ai/mlnode:3.0.14-cu129) plus the A100 env-var caveat.
• Optional: PoC delegation and refusal — added a copy-paste delegation example for MiniMax.

All vLLM argument sets are taken verbatim from the JSON files added in gonka-ai/gonka#1260 (deploy/join/node-config-minimax-{A100,H100,H200,B200}.json), so docs and repo stay in sync.

Test plan

• Render the page locally (mkdocs serve) and verify all 6 new tabs render correctly under "Edit Inference Node Description for the Server".
• Verify the new "MiniMax M2.7" tab appears under "Pre-download Model Weights to Hugging Face Cache (HF_HOME)".
• Verify the new MiniMax row appears in both the Supported models and Proposed Hardware Configuration tables.
• Verify the new "Reference deploy configs in the repo" tip box renders.
• Spot-check that the inline node-config.json blocks match the JSON files in gonka-ai/gonka:deploy/join/node-config-minimax-*.json byte-for-byte (modulo whitespace/quoting style).

Made with Cursor