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

download-models.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+set -e
+cd ~/npu
+URLS=(
+  "Codestral-22B-v0.1-Q4_K_M.gguf|https://huggingface.co/bartowski/Codestral-22B-v0.1-GGUF/resolve/main/Codestral-22B-v0.1-Q4_K_M.gguf"
+  "gpt-oss-20b-mxfp4.gguf|https://huggingface.co/ggml-org/gpt-oss-20b-GGUF/resolve/main/gpt-oss-20b-mxfp4.gguf"
+  "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf|https://huggingface.co/bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF/resolve/main/DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf"
+  "starcoder2-15b-instruct-v0.1-Q4_K_M.gguf|https://huggingface.co/bartowski/starcoder2-15b-instruct-v0.1-GGUF/resolve/main/starcoder2-15b-instruct-v0.1-Q4_K_M.gguf"
+)
+for entry in "${URLS[@]}"; do
+  fname="${entry%%|*}"
+  url="${entry##*|}"
+  if [[ -f "$fname" && $(stat -c%s "$fname") -gt 1000000000 ]]; then
+    echo "[$(date +%H:%M:%S)] SKIP $fname (already $(du -h "$fname" | cut -f1))"
+    continue
+  fi
+  echo "[$(date +%H:%M:%S)] START $fname"
+  wget -c -O "$fname" "$url" 2>&1 | tail -5
+  echo "[$(date +%H:%M:%S)] DONE  $fname ($(du -h "$fname" | cut -f1))"
+done
+echo "[$(date +%H:%M:%S)] ALL DONE"
+ls -lh ~/npu/*.gguf