Drop protobuf pin so note-seq can install

requirements.txt

@@ -1,9 +1,7 @@
 # Extra packages for a Custom Inference Endpoint.
 # Do not pin torch/transformers/accelerate — the toolkit image already has them.
-# MelGAN in this pipeline is ONNX, so onnxruntime is required even on CPU (intel-spr).
+# Do not pin protobuf: note-seq wants protobuf>=4, and a <4 pin makes pip skip the lot.
 diffusers>=0.29.0,<0.35.0
-note-seq==0.0.5
-pretty_midi==0.2.10
-onnx==1.16.2
-onnxruntime==1.19.2
-protobuf>=3.20,<4
+note-seq
+pretty_midi
+onnxruntime