Bidirectional checkpoint conversion

[hlky]

What does this PR do?

PR adds bidirectional checkpoint conversion.

e.g.

from diffusers.loaders.conversion import get_conversion

conversion = get_conversion("FluxTransformer2DModel", config)
original = conversion.to_original(diffusers_state)
restored = conversion.to_diffusers(original)

Core implementation in

src\diffusers\loaders\conversion\core.py
src\diffusers\loaders\conversion\io.py
src\diffusers\loaders\conversion\transforms.py

Per model conversion code under src\diffusers\loaders\conversion\*.py

scripts/convert_* refactored to scripts/recipes/*.py - these are "pipeline level", whereas scripts/convert_checkpoint.py is "model level", convert_checkpoint works both ways.

FromOriginalModelMixin added where appropriate.

single_file_utils.py also refactored to use loaders/conversion

Motivation: I needed something similar for DinoML.

Fixes #10076

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[hlky]

cc @yiyixuxu @sayakpaul

[sayakpaul]

Cc: @DN6