Ollama-integraatio: GPU-inferenssi NVIDIA/AMD/Apple, ei Candle-rajoitteita

- docker-compose: Ollama-container GPU:lla + persistent volume malleille
- native-node: Candle poistettu, kutsuu Ollaman HTTP API:a (async)
- Dockerfile: yksinkertaistettu, ei CUDA SDK:ta (Ollama hoitaa GPU:n)
- Tukee kaikkia malleja: qwen2.5-coder:1.5b/3b/7b/14b/32b
- OLLAMA_MODEL ympäristömuuttujalla vaihdetaan malli
- kpn models näyttää Ollama-mallit nopeustiedoilla

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

network-poc/native-node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "native-node"
-version = "0.1.0"
+version = "0.2.0"
 edition = "2024"
 
 [dependencies]
@@ -12,10 +12,6 @@ serde_json = "1.0"
 sysinfo = "0.30"
 nvml-wrapper = "0.10"
 wgpu = "24"
-candle-core = { version = "0.8", features = ["cuda"] }
-candle-nn = "0.8"
-candle-transformers = "0.8"
-hf-hub = "0.4"
-tokenizers = "0.19"
+reqwest = { version = "0.12", features = ["json"] }
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }