[core] reuse AttentionMixin for compatible classes

[sayakpaul]

What does this PR do?

Many models use "# Copied from ..." implementations of attn_processors and set_attn_processor. They are basically the same as what we have implemented in

diffusers/src/diffusers/models/attention.py

Line 39 in 693d8a3

class AttentionMixin:

This PR makes those models inherit from AttentionMixin and removes the copied-over implementations.

I decided to leave fuse_qkv_projections and unfuse_qkv_projections out of this PR because some models don't have attention processors implemented in a way that would make this seamless. But the methods removed in this PR should be very harmless.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.