Error: Could not infer shapes: unsupported: Reshape with shape containing a -1 element

[BinaryWarlock]

Describe the bug
Error: Could not infer shapes: unsupported: Reshape with shape containing a -1 element

To Reproduce
Using onnx-clip model which you can grab at https://lakera-clip.s3.eu-west-1.amazonaws.com/clip_image_model_vitb32.onnx (about 336 MiB)
$ nnx prepare -i data/clip_image_model_vitb32.onnx data/clip_image_model_vitb32.onnx.prepared --set image_batch_size=1

Expected behavior
It to prepare the model

[valpackett]

Same with these ultraface models if discarding shapes:

RUST_LOG=info nnx prepare version-RFB-320_simplified.onnx version-RFB-320_simplified.x.onnx --discard-shapes
RUST_LOG=info nnx prepare version-RFB-320_simplified.x.onnx version-RFB-320_simplified.w.onnx --infer-shapes
…
Error: Could not infer shapes: unsupported: Reshape with shape containing a -1 element

Running them (without discarding) on GPU results in

Error: backend execution error: GPU model error: compiling node 'Softmax' failed: the value '2' is invalid for attribute 'axis' (opset version 9)

pretty sure that's due to the Reshape as well.

---

Looks like this is how to handle it:

https://github.com/sonos/tract/blob/0956a664762b123b6db81cd6618ebbf13acb3234/core/src/ops/change_axes.rs#L899-L907