WIP: Custom operators

[decahedron1]

Allows custom operator kernels to be written and used entirely in Rust. See here for an example of the API.

TODO

• Documentation
• Better example (can anyone help with this?)
• Add GPU compute streams to kernel context
• Implement GetKernelAttribute for all supported attribute types
• Introduce a proc macro to make declaring operators less verbose, something like:

#[ort::operator("GroupNorm")]
fn group_norm(
    #[input]
    x: &Tensor<f32>,
    #[attribute(default = 1e-6)]
    epsilon: f32,
    #[output]
    y: &mut Tensor<f32>
) {
    ...
}

[codecov]

Codecov Report

Attention: 182 lines in your changes are missing coverage. Please review.

Comparison is base (cc40e6b) 50.68% compared to head (27e8167) 44.48%.
Report is 7 commits behind head on main.

❗ Current head 27e8167 differs from pull request most recent head 6508f07. Consider uploading reports for the commit 6508f07 to get more accurate results

Files	Patch %	Lines
src/operator/bound.rs	0.00%	66 Missing ⚠️
src/value.rs	7.89%	35 Missing ⚠️
src/operator/io.rs	0.00%	28 Missing ⚠️
src/operator/mod.rs	0.00%	19 Missing ⚠️
src/operator/kernel.rs	0.00%	17 Missing ⚠️
src/error.rs	0.00%	8 Missing ⚠️
src/lib.rs	50.00%	5 Missing ⚠️
src/session/mod.rs	20.00%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #156      +/-   ##
==========================================
- Coverage   50.68%   44.48%   -6.21%     
==========================================
  Files          15       19       +4     
  Lines        1241     1423     +182     
==========================================
+ Hits          629      633       +4     
- Misses        612      790     +178     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.