feat: add /gpu-monitor endpoint for live nvidia-smi streaming

Streams nvidia-smi dmon output over HTTP for real-time GPU utilization
monitoring during benchmark runs.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

app.py

@@ -926,5 +926,36 @@ async def benchmark_endpoint(
                 pass
 
 
+@app.get("/gpu-monitor")
+async def gpu_monitor_endpoint(duration: int = 180, interval: int = 1):
+    """Stream nvidia-smi dmon output for the given duration.
+
+    Usage: curl 'http://.../gpu-monitor?duration=180&interval=1'
+    Run this in one terminal while /benchmark runs in another.
+    """
+    import subprocess
+
+    async def _stream():
+        proc = subprocess.Popen(
+            ["nvidia-smi", "dmon", "-s", "u", "-d", str(interval)],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.STDOUT,
+            text=True,
+        )
+        try:
+            elapsed = 0
+            for line in proc.stdout:
+                yield line
+                if interval > 0:
+                    elapsed += interval
+                    if elapsed > duration:
+                        break
+        finally:
+            proc.terminate()
+            proc.wait()
+
+    return StreamingResponse(_stream(), media_type="text/plain")
+
+
 if __name__ == "__main__":
     uvicorn.run("app:app", host="0.0.0.0", port=7860, reload=False)