Backend/model integration

[conorato]

Summary

This PR includes:

• changes to use the downloaded model from our storage in order to make predictions
• code to download the hidden markov model (hmm) parameters from our storage
• modifications as to how we check if the model is up to date. We now also check if the latest modification was made after the time the model or matrix was modified in the storage. What lead to this modification is that the files containing the parameters for the HMM model will most likely have the same size. Indeed, even if we retrain and change the parameters, the matrices or vector will have the same dimensions (either N_STAGE x N_STAGE, or N_STAGE elements) and the float size of each parameter will stay the same.
• Fixed a bug in the preprocessing steps, which led to all W predictions. Our training dataset is composed of raw EEG stored in EDF files. This file format stores, by default, amplitudes in Volts, and this scale was not changed afterwards. Thus, the prediction pipeline expects data to be in V, not in uV (which was the case in the backend before).

To verify

We should check if the generated server can be executed with and without the onnxruntime's dependancy, OpenMP.

[conorato]

[1.1.6.1] Classification sur le serveur

[abelfodil]

Quelques commentaires par rapport à l'achitecture du modèle