Generalize SwiGLU related python code

[warpuv]

🚀 Feature

Generalize SwiGLU related python code. Create base classes and generalized functions to reuse by SwiGLU and other GLU-like activation functions that could be implemented in the future.

Motivation

This and some other changes allow for other similar activation functions (like GeGLU) to be implemented using the same codebase.

Additional context

I plan to add GeGLU with fused ops and packed weights implementation based on the SwiGLU code. This change is necessary to avoid code duplication.