Add muon and flash-muon optimizer

[kadirnar]

Feature request

Muon: https://github.com/KellerJordan/Muon
Flash-Muon: https://github.com/nil0x9/flash-muon
Paper: https://arxiv.org/pdf/2502.16982

Motivation

An effective optimizer method to further accelerate LLM training. The Muon team has recently proven the importance of the muon optimizer in the LLM models they released.

Your contribution

I want to add the transformers library to this optimization.