Add support for the normalized and orthogonal optimizers with matrix sign

Is your feature request related to a problem? Please describe.
The optimizer for normalized layers is introduced by Franz Cesista in https://leloykun.github.io/ponder/steepest-descent-stiefel/#6-bonus-a-muon-like-optimizer-for-the-embedding-and-unembedding-layers
and the optimizer for stiefel is introduced by Jianlin Su in https://kexue.fm/archives/11221

Further, Tilde has introduced a set of optimizers relaxing the strong constraints: https://www.tilderesearch.com/vignettes/gram-space

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Start by reading the linked Muon-like, Stiefel, and Gram-space optimizer references to determine the intended normalized and orthogonal behaviors. Then inspect the repository's existing optimizer entry points and tests, which are not named in the issue; done means the requested matrix-sign-based optimizers are implemented with coverage for their expected behavior.