Support for opset 14

[Swopper050]

Closes #216

[Copilot]

Pull Request Overview

This PR adds support for ONNX opset 14 across several operators. Key changes include:

• Addition of version 14 operator constructors and related type constraints for multiple ops (trilu, sub, rnn, relu, mul, lstm, identity, gru, div, cumsum, and add).
• Introduction of a new layout attribute and associated reshape logic in RNN, LSTM, and GRU operators to support batch-first tensor layouts.
• Updates to corresponding test files and documentation to reflect opset 14 support.

Reviewed Changes

Copilot reviewed 25 out of 25 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
ops/trilu/*	Added tests and implementation updates for opset 14 support
ops/sub/versions.go	Added version 14 operator constructor and type constraints
ops/rnn/*	Extended operator and tests with layout support and version 14
ops/relu/versions.go	Added version 14 support and new type constraints
ops/mul/versions.go	Added version 14 operator constructor and type constraints
ops/lstm/*	Extended operator and tests with layout support and version 14
ops/identity/versions.go	Added version 14 operator constructor
ops/gru/*	Extended operator and tests with layout support and version 14
ops/div/versions.go	Added version 14 operator constructor and type constraints
ops/cumsum/versions.go	Added version 14 operator constructor
ops/add/versions.go	Added version 14 operator constructor and type constraints
README.md	Updated documentation to reflect support for opset 14

Comments suppressed due to low confidence (2)

ops/rnn/rnn_test.go:407

• Ensure that the package 'math/rand' is imported explicitly in this test file if it is not already, as the use of rand.New requires it.

r := rand.New(rand.NewSource(13))

ops/rnn/rnn.go:111

• Reshaping the input tensor X directly may result in unintended side effects if the original tensor is reused elsewhere. Consider cloning the tensor before reshaping.

err := X.Reshape(seqLength, batchSize, inputSize)