FTAG: Adding support in Athena for Gn1 Aux tasks

Can support multiple output types for the onnx models -

• float
• vec<float>
• vec<char>

---

Introducing GNNConfig::Config

• The goal is to eventually get rid of lwtnn dependency.
• Right now, we look at GNNConfig::Config for creating the decorators. (Will include configs for accessors as well in future)

---

• Decorators are created from reading the outputs of the onnx model and not from the metadata as before
• (In future, it can be extended for the accessors as well)

---

Changes to the model output format

• Previous models were returning pb, pc, pu scores as vec<float> (of length 1). We had to extract the first element and then decorate it as a float
• The new models return the exact type it needs to be decorated as. so float for pb, vec<char> for track_class etc.
• Everything is figured out from the onnx model directly, and it doesn't require to read the metadata for it anymore.

---

Backward compatibility is assured!