[Schedulers] Add methods for conversion between noise and denoised predictions

Guidance methods like APG (#9585 #9626) require conversion between noise and denoised predictions. The computation of these predictions varies across schedulers and depends on `prediction_type`. Currently, this is embedded in `step` functions of the schedulers which complicates implementation of APG.

Adding separate methods for conversion between noise and denoised predictions would simplify implementation and allow easier experimentation.

Start by reviewing the schedulers' step functions and their prediction_type handling, then read the APG discussions in issues #9585 and #9626. Done means the conversion logic is exposed as separate scheduler methods and supports the prediction variants needed by the guidance methods.

Rédigé par le modèle d'indexation à partir du texte de l'issue.