[CK-Tile] add persistent async input scheduler parameters to kernel device-side and host-side args

[tenpercent]

Summary

• Add PersistentAsyncInputScheduler struct for signal-based synchronization in persistent GEMM kernel
• Implement signal wait logic using modulo wraparound (matching PyTorch's AsyncMM)
• Add wait_eq_wave method with power-saving sleep for efficient busy-waiting
• Add GTest unit tests covering all layout combinations

The future client side here

Description

This PR implements signal-based synchronization for the persistent GEMM kernel, enabling async input streaming use cases where input data arrives in chunks.

Key Features

1. PersistentAsyncInputScheduler struct (include/ck_tile/core/utility/persistent_async_input_scheduler.hpp):

• tiles_per_chunk_m - number of M tiles per chunk
• chunk_signals - pointer to device signal array (one per chunk)
• tile_idx_pivot_m - pivot offset for chunk index calculation
• num_chunks - number of chunks for modulo wraparound

2. Kernel signal wait logic (include/ck_tile/ops/gemm/kernel/universal_gemm_kernel.hpp):

• Conditional check for chunk_signals != nullptr
• Uses modulo wraparound like PyTorch's AsyncMM to avoid out-of-bounds access:

  chunk_idx = ((iM + tile_idx_pivot) / tiles_per_chunk) % num_chunks
  

• Uses workgroup_barrier::wait_eq_wave() with __builtin_amdgcn_s_sleep(1) for power-efficient waiting
• IsSupportedArgument validation ensures tiles_per_chunk > 0 and num_chunks > 0

3. Power-optimized wait_eq_wave (include/ck_tile/core/arch/workgroup_barrier.hpp):

• Only lane 0 polls memory, broadcasts result via __shfl
• Uses __builtin_amdgcn_s_sleep(1) to reduce power consumption during busy-wait
• 63 of 64 threads in the wave sleep during the polling loop

4. Example and Tests:

• Example with test_async=1 flag demonstrating async input scheduling
• GTest unit tests covering RowRow, RowCol, ColRow, ColCol layouts

Checklist

Please put an x into the boxes that apply. You can also fill these out after creating the PR. If you're not sure, please don't hesitate to ask.

• I have added tests relevant to the introduced functionality, and the unit tests are passing locally
• I have added the test to REGRESSION_TESTS list defined at the top of CMakeLists.txt in tests/CMakeLists.txt, IF the test takes more than 30 seconds to run.
• I have added inline documentation which enables the maintainers with understanding the motivation
• I have removed the stale documentation which is no longer relevant after this pull request
• (If this change is user-facing) I have added release notes which provide the end users with a brief summary of the improvement from this pull request
• I have run clang-format on all changed files
• Any dependent changes have been merged

Discussion

If this is a relatively large or complex change, feel free to start a discussion by explaining why you chose the solution you did and what alternatives you considered

[ThomasNing]

@tenpercent Please fix the clang format.