The scale of timesteps in the text modeling examples

Hi! Really enjoying the code and the paper. These are the most comprehensive resources on flow matching.

I noticed that [the text modeling example](https://github.com/facebookresearch/flow_matching/blob/main/examples/text/model/transformer.py) might contain a subpar encoding of timesteps. In short, it uses the default `max_period` parameter from the GLIDE repository, which is a diffusion model with discrete timesteps in the range {0, 1, 2, ...}. However, as I understand correctly, in your example, all timesteps are sampled from the [0, 1] time horizon. See [the similar issue](https://github.com/willisma/SiT/issues/24) that I've opened in the SiT repository. I think the most elegant solution is to rescale the timesteps similar to how [Flux](https://github.com/black-forest-labs/flux/blob/main/src/flux/modules/layers.py#L28) handles them.

Let me know if I am misunderstanding something, as I am just on my way of learning the flow matching framework.