Feature request: Lumina2 architrcture support

[stduhpf]

Lumina-Image-2.0 is a 2B flow-based diT. Image quality seems pretty good at first glance, and it's able to generate some text.
https://huggingface.co/Alpha-VLLM/Lumina-Image-2.0

The text "encoder" used is gemma2-2B (autoregressive LLM), like Nvidia Sana.
It uses a 16 channel VAE like Flux and SD3, so this would make it easier to implement than sana.