fix: use complete URL env var for vLLM API base

- Replace GPU_TAILSCALE_IP interpolation with GPU_VLLM_API_URL
- LiteLLM requires full URL in api_base with os.environ/ syntax

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

ai/compose.yaml

@@ -102,6 +102,7 @@ services:
       LITELLM_MASTER_KEY: ${AI_LITELLM_API_KEY}
       DATABASE_URL: postgresql://${AI_DB_USER}:${AI_DB_PASSWORD}@ai_postgres:5432/litellm
       GPU_TAILSCALE_IP: ${GPU_TAILSCALE_IP}
+      GPU_VLLM_API_URL: ${GPU_VLLM_API_URL}
       # LITELLM_DROP_PARAMS: 'true'  # DISABLED: Was breaking streaming
       NO_DOCS: 'true'
       NO_REDOC: 'true'

ai/litellm-config.yaml

@@ -33,7 +33,7 @@ model_list:
   - model_name: qwen-2.5-7b
     litellm_params:
       model: hosted_vllm/openai/qwen-2.5-7b  # hosted_vllm/openai/ for vLLM via orchestrator
-      api_base: http://os.environ/GPU_TAILSCALE_IP:9000/v1  # RunPod GPU via Tailscale
+      api_base: os.environ/GPU_VLLM_API_URL  # RunPod GPU via Tailscale
       api_key: dummy
       rpm: 1000
       tpm: 100000
@@ -45,7 +45,7 @@ model_list:
   - model_name: llama-3.1-8b
     litellm_params:
       model: hosted_vllm/openai/llama-3.1-8b  # hosted_vllm/openai/ for vLLM via orchestrator
-      api_base: http://os.environ/GPU_TAILSCALE_IP:9000/v1  # RunPod GPU via Tailscale
+      api_base: os.environ/GPU_VLLM_API_URL  # RunPod GPU via Tailscale
       api_key: dummy
       rpm: 1000
       tpm: 100000