[Feature Request] check async copy is used for auto T.Pipelined

### Required prerequisites

- [x] I have searched the [Issue Tracker](https://github.com/tile-ai/tilelang/issues) that this hasn't already been reported. (comment there if it has.)

### Motivation

Tilelang will use ldg.256+`tl::cp_async_commit()`+`tl::cp_async_wait`, when auto vectorized 256 bits vector, T.Pipelined is opened, and tma&warpspecialized are disabled. The expected behavior is to use 128bits `cp.async.copy`.

### Solution

_No response_

### Alternatives

_No response_

### Additional context

_No response_

Start by locating the auto-vectorization and T.Pipelined code paths that emit ldg.256, tl::cp_async_commit(), and tl::cp_async_wait when TMA and warp specialization are disabled. Trace the generated async-copy width and verify that the 128-bit cp.async.copy behavior is selected instead of the current 256-bit path.

Written by the indexing model from the issue text.