Use GitHub orpheus-tts (supports max_model_len) to fix OOM on Jetson

main.py

@@ -326,9 +326,11 @@ async def startup():
     print("Loading Orpheus model (this may take a moment)...")
     from orpheus_tts import OrpheusModel
     
-    # Note: PyPI orpheus-speech 0.1.0 uses simpler API
-    # model_name can be "medium-3b" or full HF path
-    model = OrpheusModel(model_name=ORPHEUS_MODEL)
+    # GitHub version supports max_model_len for memory control
+    model = OrpheusModel(
+        model_name=ORPHEUS_MODEL,
+        max_model_len=MAX_MODEL_LEN
+    )
     
     print("✓ Orpheus model loaded successfully")