Refactor effort constraints

Changes:

• Add the use of effort constraint parameters in federated evaluation rounds.
• Add fields in EvaluationRequest and EvaluationReply messages wrapping these parameters.
• Add utils to specify constraints under declearn.main.utils, that lighten the code implementing the constrained training and evaluation rounds.

Notes:

• This code paves the way for future constraint features, such as measuring and limiting the energy cost of computations.
• Further refactoring may be used to:
  • group together constraint parameters into a dict (in methods and messages)
  • limit code redundancies as to instantiating constraints
  • turn train_under_constraints and evaluate_under_constraints into generic functions that may be re-used when implementing alternative federated learning orchestrating classes