feat: add Atlas Inference Engine provider detection (closes #674)

Add Atlas (pure Rust+CUDA inference engine by Avarok) to the provider
registry and local server detection.

Changes:
- agent/model_metadata.py:
  - _URL_TO_PROVIDER: add localhost:8888 -> atlas mapping
  - detect_local_server_type: add Atlas detection via /health endpoint
    (checks for engine=atlas in response) and /v1/models fallback
  - Update docstring to include atlas in return type

Atlas serves OpenAI-compatible API at localhost:8888/v1 from a 2.5 GB
Docker image. 3x faster than vLLM with zero Python dependencies.

Quick start:
  docker pull avarok/atlas-gb10:alpha-2.8
  docker run -d --gpus all --ipc=host -p 8888:8888 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    avarok/atlas-gb10:alpha-2.8 serve \
    Sehyo/Qwen3.5-35B-A3B-NVFP4 --speculative

Hermes config (config.yaml):
  provider: custom
  base_url: http://localhost:8888/v1

agent/model_metadata.py

@@ -233,11 +233,13 @@ _URL_TO_PROVIDER: Dict[str, str] = {
     "api.deepseek.com": "deepseek",
     "api.githubcopilot.com": "copilot",
     "models.github.ai": "copilot",
-    "api.fireworks.ai": "fireworks",
-    "opencode.ai": "opencode-go",
-    "api.x.ai": "xai",
-    "api.xiaomimimo.com": "xiaomi",
-    "xiaomimimo.com": "xiaomi",
+    \"api.fireworks.ai\": \"fireworks\",
+    \"opencode.ai\": \"opencode-go\",
+    \"api.x.ai\": \"xai\",
+    \"api.xiaomimimo.com\": \"xiaomi\",
+    \"xiaomimimo.com\": \"xiaomi\",
+    # Atlas Inference Engine — pure Rust+CUDA, OpenAI-compatible at :8888
+    \"localhost:8888\": \"atlas\",
 }
 
 
@@ -302,10 +304,12 @@ def is_local_endpoint(base_url: str) -> bool:
     return False
 
 
+import json
+
 def detect_local_server_type(base_url: str) -> Optional[str]:
     """Detect which local server is running at base_url by probing known endpoints.
 
-    Returns one of: "ollama", "lm-studio", "vllm", "llamacpp", or None.
+    Returns one of: "ollama", "lm-studio", "vllm", "llamacpp", "atlas", or None.
     """
     import httpx
 
@@ -355,6 +359,25 @@ def detect_local_server_type(base_url: str) -> Optional[str]:
                         return "vllm"
             except Exception:
                 pass
+            # Atlas Inference Engine: OpenAI-compatible, custom /health
+            # Atlas returns {"status":"ok","engine":"atlas"} on /health
+            try:
+                r = client.get(f"{server_url}/health")
+                if r.status_code == 200:
+                    try:
+                        data = r.json()
+                        if data.get("engine") == "atlas" or "atlas" in str(data).lower():
+                            return "atlas"
+                    except Exception:
+                        pass
+                # Fallback: check /v1/models for Atlas-specific response
+                r = client.get(f"{server_url}/v1/models")
+                if r.status_code == 200:
+                    data = r.json()
+                    if "atlas" in json.dumps(data).lower():
+                        return "atlas"
+            except Exception:
+                pass
     except Exception:
         pass