Partial output tiles serialize stores through one lane

Problem

The recurrence kernel uses TMA for full 16-row output tiles, but falls back to a manual scalar copy for the final partial tile so it does not overwrite the next sequence.

That fallback currently executes only on the STORE warp lane selected by elect_one_sync(). With D=128, a tail of 15 rows therefore makes one lane perform:

15 rows * 128 columns = 1,920 scalar stores

The indexing is boundary-safe, but the entire tail copy is serialized.

Affected path

csrc/smxx/fwd_kernel2.cuh, in the actual_len < CHUNK output-store branch.

Desired outcome

• Retain elected-lane ownership of output-pipeline wait and release operations.
• Retain elected-lane ownership of full-tile and final-state TMA stores.
• Distribute only the partial-tile scalar copy across the STORE warp.
• Synchronize the warp before the shared-memory stage is released.
• Cover tail lengths 1 through 15 at adjacent variable-length sequence boundaries.

A focused benchmark can be used to quantify the effect separately; the ownership and serialization issue is visible directly in the current kernel.

Start in csrc/smxx/fwd_kernel2.cuh at the actual_len < CHUNK output-store branch and inspect the existing STORE warp lane ownership and synchronization. Distribute only the partial-tile scalar copy across the STORE warp while preserving elected-lane pipeline and TMA operations. Done means tail lengths 1 through 15 are covered at adjacent variable-length sequence boundaries and the shared-memory stage is synchronized before release.