Initial commit: Boltzmann LLM proxy with compression + classifier middleware

- llm-proxy.py v4: aggregator with classify/compress endpoints, gemma4 backend
- classifier-server.py: Llama 3.2 1B query complexity classifier
- compress-server.py: token-budget compression middleware
- start-proxy.sh / start.sh: launcher scripts

launch-qwen35-npu.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+# Manual launcher for the NPU Qwen 3.5 9B llama-server backend.
+# Used for ad-hoc testing; the systemd user unit
+# llama-server-qwen35-npu.service is the production path.
+set -e
+ulimit -n 65536
+exec /home/mfritsche/src/rk-llama.cpp/build/bin/llama-server \
+  -m /home/mfritsche/models/Qwen3.5-9B-Q8_0.gguf \
+  --alias qwen3.5-9b-npu \
+  --reasoning off \
+  -c 4096 -t 4 \
+  --host 0.0.0.0 --port 8086