Add scale factor to the LDM training tutorials

[Warvito]

One possible way to the LDM training fails is due to the inappropriate range values that the latent representation of the autoencoder might create. This issue is more crucial in the upscalers, but it could also affect in the image generation. Would be okay to include an extra step in the tutorials about LDM, calculating the scale factor (similar to what is already discussed in

GenerativeModels/tutorials/generative/2d_super_resolution/2d_stable_diffusion_v2_super_resolution.py

Line 302 in fd362dc

# ## Rescaling factor

)? What do you think @SANCHES-Pedro @JessyD ?

[virginiafdez]

I can help with this

[virginiafdez]

@Warvito Pushed. Could you quickly check the added text on the tutorials and run them?

[virginiafdez]

@Warvito Pushed changes to 2d_ldm.
I ran it for a few epochs and it works, but I can't run for more (it times out).
Could you check it and run it?

[Warvito]

mrged in #271