ONNX assets for the backend's eye state, smile and gaze analysis,
fetched by POST /v1/models/assets/downloads.
face_mesh_v2.onnx— MediaPipe Face Mesh V2 (478 landmarks),
Apache-2.0face_blendshapes_v2.onnx— MediaPipe Blendshape V2 (52
expression coefficients), Apache-2.0
Converted from Google's face_landmarker.task with tf2onnx; see
server-rs/scripts/export_face_state.py and
.github/workflows/model-assets-face-state.yml.