made DDPM compatible with UNet2DConditionModel

[Mafaz03]

Summary

This PR addresses compatibility issues when using the UNet2DConditionModel with the DDPM pipeline. The issue arises because the UNet2DConditionModel requires encoder_hidden_states, which is not currently handled by the DDPM pipeline.

Problem Statement

Models such as UNet2DConditionModel [https://huggingface.co/docs/diffusers/en/api/models/unet2d-cond]
requires encoder_hidden_states that doesnt work well with DDPM pipeline as it doesnt have it as input parameter.

Fixes

• Updated the DDPM pipeline’s __init__ method to accept encoder_hidden_states as an optional parameter.
• Modified subsequent methods in the pipeline to pass encoder_hidden_states to the model during inference and training.
• Ensured backward compatibility by keeping encoder_hidden_states optional.