Organize dependencies

• Declare dependencies for each model as an extra. This means that users would only have to download dependencies for the model they want to use by including its name in the installation command.

• Add pytorch index as a secondary source for poetry. The torch package on PyPI is the CUDA version that installs proprietary NVIDIA packages, as pointed out in T325349. Since we don't need CUDA, the source points to the cpu index. Also, because torch already exists on PyPI, this extra index should not make the package more susceptible to dependency confusion attacks.