feat: add teachable machine with custom trained model

[dido18]

Motivation

I discovered the amazing teachable machine project.
It allows training a model (via tensorflow.js) via a web browser into 3 different fields: images. sound, pose detection

Advantages with respect to edge impulse: runs ona web browser, the trainer is for kids, lower latency.

Changes

• expose an HTTP endpoint to download the model. A model is a folder with the following files:

├── metadata.json  
├── model.json
└── weights.bin

• extension to download the model and use it