Update MAEST models

[palonso]

This PR updates Essentia models's website and algorithms to accommodate the recently added MAEST-related models. These are:

• V2 version of the MAEST feature extractors
• V1 classification heads for the genre_discogs400 and genre_discogs519 task

The main motivation for the V2 version of the MAEST models was:

• Fixing issue related to the embedding layer retrieval, since for the V1 models the index was incorrect (e.g., StatefulPartitionalCall:4 did not correspond to the embeddings of the 4th layer of the model).
• Enhancing inference times thanks to the conversion and optimization from ONNX via ONNX2TF.

The conversion code is available in this repository.

The contributions of this PR are:

• Update the default TensorfowPredictMAEST input and output parameters to fit the new MAEST model names.
• Fix incorrect patch size for the 20s version of MAEST in TensorfowPredictMAEST.
• Update MAEST embedding extraction example scripts.
• Create genre_discogs400 and genre_discogs519 example scripts.
• Improve the readability of the script in charge of generating the model example scripts
• Add scripts to batch-execute all the example scripts