add fondations for LLM

[tharvik]

merging LLM (from #619) requires some reshaping of the code, here goes some changes splitted from the original PR.

• add a Model abstracting away tensorflow specifics
  • still incomplete as it input/output tf.Tensors for now, and one can extract the tf.LayersModel
  • remove Task.training.modelCompileData as it's now contained inside the models.TFJS
• add typesafe event emitter to safely register/call callbacks
• rename Task.taskID to Task.id
• while at it: drop unused training_functions, update axios, use async file read

[JulienVig]

For what it's worth very good stuff! I only left comments for what I didn't understand but it's a very nice refactoring

[tharvik]

For what it's worth very good stuff! I only left comments for what I didn't understand but it's a very nice refactoring

héhé, thanks 😄

followup of fixes from your remarks #645