burn-import: add some tests for ConstantNode

[jameshiew]

Pull Request Template

Checklist

• Confirmed that run-checks all script has been executed.
• Made sure the book is up to date with changes in this PR.

Related Issues/PRs

I've been trying to implement OneHot ONNX op (#1714) in a WIP draft branch. The ONNX model ends up containing a constant integer vector values=[0, 1] used by the OneHot op, this vector was causing issues when trying to test the model. I looked at ConstantNode and these are the tests so far I could get working while investigating.

Issues for ConstantNode
#2624 - constant tensors aren't populated with values
#2625 - generated code for const int tensors doesn't compile

Changes

• added a helper method ConstantNode::tensor_ty_tokens for tests, but this PR otherwise shouldn't be changing how ConstantNode currently works
• add codegen tests for ConstantNode (i32/64 + f32/64 scalar, tensors)
• add ONNX model tests for i32/64 + f32/64 scalars - implicitly testing by adding the constant

Testing

Ran added tests

cargo xtask check all
cargo nextest run --manifest-path crates/burn-import/Cargo.toml
cargo nextest run --manifest-path crates/burn-import/onnx-tests/Cargo.toml

I checked the .onnx models contain the expected scalar constants using Netron

Screenshots

[jameshiew]

I'm not sure if the addition is coercing f64 -> f32 somewhere (and i32 -> i64 below). I wasn't sure how to get PyTorch to just forward the constant by itself so these tests are adding the constant to the input