VAE or VQ-VAE is needed

**Is your feature request related to a problem? Please describe.**
Currently, the timm library lacks implementations for Variational Autoencoder (VAE) and Vector Quantized VAE (VQ-VAE) models. Users looking to utilize these autoencoder architectures may find it inconvenient to implement them from scratch or integrate external implementations into their projects.

**Describe the solution you'd like**
I would like to request the addition of Variational Autoencoder (VAE) and Vector Quantized VAE (VQ-VAE) models to the timm library. This would involve creating modules for these autoencoder architectures, ensuring they adhere to the existing timm standards for simplicity and compatibility.

**Describe alternatives you've considered**
Users can currently implement VAE and VQ-VAE models from scratch or use external implementations from other libraries such as diffusener. However, having native support for these models in the timm library would provide a more streamlined and integrated experience for users.

No files, tests, or entry points are named. Start by reviewing existing timm model modules and the library's conventions for simplicity and compatibility, then define the required VAE and VQ-VAE scope; done would mean both architectures are implemented as native timm models and meet those standards.

Written by the indexing model from the issue text.