Reorders stride parameter assignments

[LoserCheems]

Reorganizes the order of stride parameter assignments to group related parameters together for better code readability and maintainability.

Adds missing column stride assignments for zoh and active_mask tensors.

[Copilot]

Pull Request Overview

Reorganizes the assignment order of output strides and introduces missing column-stride parameters for zoh and active_mask tensors to improve readability and ensure all tensor strides are captured.

• Reordered o_row_stride next to other row-stride assignments.
• Added zoh_col_stride and active_mask_col_stride assignments.
• Grouped related stride assignments by “row”, “head”, and now “col”.

Comments suppressed due to low confidence (2)

csrc/flash_api.cpp:85

• New column-stride assignments for zoh and active_mask have been added but aren’t covered by existing tests. Please add unit tests to verify these stride values.

    params.zoh_col_stride = zoh.stride(-1);

csrc/flash_api.cpp:85

• Update the set_params_fprop function header or the params struct documentation to include descriptions of the new zoh_col_stride and active_mask_col_stride fields.

    params.zoh_col_stride = zoh.stride(-1);