Fedbiomed

- fix secure aggregation vector encoding bug

- add HTTPS secure access to Fed-BioMed Node GUI
- introduce GitHub workflow/actions for CI build tests and testing/publishing documentation.
- introduce versioning for component config files, MQTT messages and breakpoints
- migrate to GitHub
- migrate `docs` source into main repository and point to https://fedbiomed.org
- fix robustness of handling secure aggregation training errors.
- fix warnings in `TorchModel.set_weights` and BatchNorm layers' handling
- fix incorrect calculation of SkLearn model weights
- fix incorrect compatibility of FedProx feature with `Model` class
- fix ordering of weights and node replies after training.

- introduce secure aggregation using Joye-Libert scheme and Shamir MPC key computation
- update MONAI and scikit-learn version used
- fix Scaffold incorrectly applying correction states
- fix incorrect Perceptron default values for scikit-learn models
- fix `Experiment.set_training_args()` not propagating updated value
- fix environment cleaning to handle configuration file content change
- fix docker wrapping scripts to restrict container account names to alphanumeric characters
- misc improve node CLI for non-interactive add of MedicalFolderDataset using a json file

- add support for `docker compose` v.2 file syntax
- fix model weights computation occurring during aggregation, by sending dataset sample size from node to researcher
- fix GUI regression failure, after merging MP-SPDZ certificate generation - such issue was freezing some web browsers
- fix incoherent tag handling: make explicit the way datasets are tagged on nodes
- fix unit tests failure, when launched from root directory, due to missing mocking facility
- fix `fedbiomed_run` error: prevent launching researcher when no config file exists
- misc improve make sure only one dataset per Node is selected during the training
- misc remove uncorrect warning about `optimizer_args` when using SKlearn training plan

- introduce Scaffold implementation for PyTorch
- introduce training based on iteration number (`num_updates`) as an alternative to epochs
- introduce provisions for including external contributors in the project
- add nightly continuous integration test of selected notebooks
- add documentations for network matrix and security model
- update image segmentation notebook to match documentation tutorial
- add round number in researcher side training progress message
- fix MedicalFolderDataset with demographics file using column 0 as subject folder key
- fix batch size display when using Opacus
- fix loss display when using FedProx
- fix default value of `batch_maxnum` to a reasonable value of 0
- fix validation with custom metrics backend code and validation example notebook
- fix image segmentation notebook typo
- misc improve MedicalFolderDataset with a reasonable default value for `demographics_transform`
- misc improve error message when dataset geometry does not meet researcher side quality check

- introduce IXI (image + CSV file) dataset support as MedicalFolderDataset
- add advanced brain image segmentation tutorial for IXI dataset
- add node side GUI support for IXI dataset
- major redesign of training plan implementation for genericity
- redesign Opacus integration with torch training plan
- implement central and local differential privacy (CDP/LDP) for Pytorch training plan
- introduce integration with FLamby FL benchmark package
- introduce node side GUI user accounts, authentication, accounts management
- introduce data loading plan functionality for dataset load-time custom view on the node side
- add data loading plan support for IXI medical folder dataset
- introduce training plan approval capability in application: researcher request, node approval CLI
- add node side GUI support for training plan approval
- introduce mini-batch support in scikit-learn training plans
- refactor scikit-learn training plans with hierarchical design
- refactor NIFTI folder dataset type for code quality and robustness
- TrainingArgs class to manage/verify training arguments on researcher side
- rename model approval as training plan approval for coherency
- add sample notebook for researcher-side filtering of datasets on minimum samples number
- obfuscate node side path to researcher for better privacy
- misc node side TinyDB database access refactor for code quality and robustness
- misc improve scikit-learn training plan dependency handling
- fix bug on training plan report of sample/percentage progress
- fix missing `fedprox_mu` parameter in training args
- fix dry run mode for pytorch training plan
- fix conda environment GLIBC version issue