Implement Unit-Scaled Maximal Update Parametrization for best LR

There is a new paper, "u-μP: The Unit-Scaled Maximal Update Parametrization" at https://arxiv.org/abs/2407.17465 that promises selection of optimal hyperparameters, like the learning rate, even when using very small datatypes like FP8

The content is too advanced for me, but I had the impression that it might be a valuable addition to kohya_ss

Start by reading the linked u-μP paper to determine the required parametrization and whether it applies to this project. Then inspect the repository's training and optimizer entry points to identify where learning-rate and datatype behavior are configured. Done means a concrete integration plan, implementation, and validation of the claimed hyperparameter behavior.

Written by the indexing model from the issue text.