Implement Secure Aggregation

This PR brings Secure Aggregation to DecLearn.

Summary

In summary, this PR brings:

• an API for SecAgg:
  • Encrypter API to encrypt cleartext values (uint, float, numpy array, declearn Vector, declearn Aggregate)
  • Decrypter API to decrypt aggregated encrypted values
  • SecaggSetupClient and SecaggSetupServer to configure and set up controllers (via some communications)
• an integration of SecAgg into FederatedServer and FederatedClient
  • server and clients must specify compatible SecAgg configurations
  • (for current protocols: clients need to have shared identity keys)
  • server can prompt clients to set encryption/decryption controllers afresh, and will do so whenever it is required
  • clients use their encrypter (and raise if one is missing) to filter and encrypt the data sent after a training or evaluation round
• Shared utils for SecAgg and its setup.
  • Quantization utils (to convert floats to unsigned integers in a given field)
  • Shamir Secret Sharing, based on Shamir, 1979
  • X3DH key-agreement protocol, from Marlinspike & Perrin, 2016
  • Utils to handle Ed25519 identity keys and load (or save) them from (to) file(s).
• 2 SecAgg protocols:
  • Masking-based SecAgg, based on Bonawitz et al., 2016
  • Sum-homomorphic Joye-Libert encryption, based on Joye & Libert, 2013

Roadmap

• Implement Joye-Libert SecAgg
• Implement X3DH protocol
• Implement Quantization utils
• Implement Masking-based SecAgg
• Abstract away the code into a common API
• Write unit tests for all SecAgg components
• Write SecAgg-counterparts to some existing messages
• Integrate SecAgg to main orchestration classes
• Write integration tests for FL with SecAgg
• Write unit tests for FederatedClient and FederatedServer, that notably cover SecAgg features

Closes #34 (closed)