feat: Complete CNN

[sdivyanshu90]

No description provided.

[Copilot]

Pull request overview

This PR adds Convolutional Neural Network (CNN) support to the Sequre secure multi-party computation framework, introducing new layer types (Conv2D, MaxPooling2D, Flatten, Dropout, BatchNormalization) and additional loss functions (binary cross-entropy, categorical cross-entropy, mean squared error) to enable training CNNs on encrypted data.

Changes:

• Added CNN layer implementations with secure computation support including Conv2D with strided convolutions, MaxPooling2D, Flatten for dimensionality reduction, Dropout for regularization, and BatchNormalization
• Extended loss functions with binary/categorical cross-entropy using Chebyshev polynomial approximations and mean squared error
• Updated Sequential model to handle heterogeneous layer types (CNN and Dense) with automatic channel/size inference
• Added ChestMNIST medical imaging example demonstrating CNN training on multi-label classification with data preparation script and comprehensive documentation

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
stdlib/sequre/stdlib/learn/neural_net/model.codon	Updated import to relative style; added layer type detection logic for automatic prev_size inference supporting CNN layers; added section comments for training methods
stdlib/sequre/stdlib/learn/neural_net/loss.codon	Added binary/categorical cross-entropy and MSE loss functions with Chebyshev approximations; updated dispatch functions with optional interval parameter; added supporting imports
stdlib/sequre/stdlib/learn/neural_net/layers.codon	Implemented Conv2D, MaxPooling2D, Flatten, Dropout, and BatchNormalization layers with forward/backward passes; added secure maximum helper function
stdlib/sequre/constants.codon	Added constants for new loss types (BCE, CCE, MSE) and updated SUPPORTED_LOSSES list
scripts/sequre-test.sh	Added helper script for running Sequre tests with proper configuration
applications/offline/chestmnist_prep.py	Added Python script to download and prepare ChestMNIST dataset as text files for Codon
applications/offline/chestmnist_cnn.codon	Added complete CNN training example for ChestMNIST with mini-batch training and evaluation
applications/offline/CHESTMNIST_CNN_README.md	Added comprehensive documentation for running the ChestMNIST CNN example
examples/local_run.codon	Removed old example file
examples/hastings.codon	Removed old benchmark file

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The command in this script interpolates the untrusted script argument $1 directly into a string executed by bash -c, which enables command injection (e.g., passing a value containing "; rm -rf /; will execute additional arbitrary commands with the caller's privileges). Because $1 is expanded into the inner shell script and only wrapped in quotes, an attacker controlling this argument can break out of the quoted context and append their own shell syntax. To fix this, avoid bash -c here and invoke the codon binary directly with $1 passed as a normal, separately-quoted argument (or build the command as an array) so that it is never re-interpreted as shell code.

Suggested change

	bash -c "CODON_DEBUG=lt $HOME/.codon/bin/codon run --disable-opt='core-pythonic-list-addition-opt' -plugin sequre \"$1\" --skip-mhe-setup 2>&1"
	CODON_DEBUG=lt "$HOME/.codon/bin/codon" run --disable-opt='core-pythonic-list-addition-opt' -plugin sequre "$1" --skip-mhe-setup 2>&1