feat: increase Llama context to 32K with 95% VRAM

For larger input requirements, increased max-model-len from 20480 to 32768.
BGE remains available but cannot run concurrently at this VRAM level.

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

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

services/vllm/config_llama.yaml

@@ -2,8 +2,8 @@ model: meta-llama/Llama-3.1-8B-Instruct
 host: "0.0.0.0"
 port: 8001
 uvicorn-log-level: "info"
-gpu-memory-utilization: 0.90
-max-model-len: 20480
+gpu-memory-utilization: 0.95
+max-model-len: 32768
 dtype: auto
 enforce-eager: false
 enable-auto-tool-choice: true