feat: add Qwen-AgentWorld-35B-A3B recipe (2× DGX Spark, TP=2)

35B total / 3B active MoE world model for agentic environment simulation.
Requires --language-model-only (no vision weights in checkpoint).
262K context, reasoning-parser qwen3.

recipes/qwen-agentworld-35b/recipe.toml

@@ -0,0 +1,25 @@
+engine = "vllm"
+name = "Qwen-AgentWorld-35B-A3B (vLLM)"
+docker_image = "ghcr.io/timothystewart6/vllm-gb10:v0.23.0-gb10.1"
+model = "Qwen/Qwen-AgentWorld-35B-A3B"
+required_nodes = 2
+tensor_parallel_size = 2
+pipeline_parallel_size = 1
+distributed_backend = "mp"
+
+# 35B total / 3B active MoE — world model for agentic environment simulation.
+# 2× DGX Spark (GB10), 128 GB shared each.
+# TP=2 → ~35 GB/node for weights, room for 262K context.
+# Requires --language-model-only (no vision weights in checkpoint).
+
+entrypoint = ["vllm", "serve"]
+
+args = """
+--port 30000
+--max-model-len 262144
+--reasoning-parser qwen3
+--language-model-only
+--trust-remote-code
+--gpu-memory-utilization 0.90
+--enable-prefix-caching
+"""