Improvement readability and comprehensibility of cast process of generate_image() in Stable Diffusion

[y-vectorfield]

According to line 238, the casting method is implemented abruptly and unnaturally.
I added this method previously in pull request #2102.
Hence, I should move this method to line 237 and add comments about the reason for the casting.

keras-cv/keras_cv/models/stable_diffusion/stable_diffusion.py

Line 238 in 925d99d

latent = ops.cast(latent, latent_prev.dtype)

[divyashreepathihalli]

changes merged

[y-vectorfield]

@divyashreepathihalli, thank you very much!