Enhancement: Refactor Loader and Storer Implementations

[lcy-seso]

Description:

• Re-implement the loader to issue cp.async directly instead of using CuTe.
• Re-implement the storer to issue the corresponding PTX for storing, avoiding CuTe.

Benefits:

This refactor will offer several advantages and help achieve a cleaner codebase. Currently, Cutlass requires its own Layout, which necessitates wrapping the parameters for Cutlass in various places. This results in code that is messy and difficult to manage. By eliminating the use of CuTe in favor of direct PTX instructions, we can streamline the implementation and improve code clarity.