update allreduce to match trtllm (#1507)

<!-- .github/pull_request_template.md -->

## 📌 Description

<!-- What does this PR do? Briefly describe the changes and why they’re
needed. -->

Updated allreduce launch config logic to match trtllm.
On llama3 concurrency=128 tp2 gen-only phase, the kernel time improved
from ~26.8us to ~9.8us.

## 🔍 Related Issues

<!-- Link any related issues here -->

## 🚀 Pull Request Checklist

Thank you for contributing to FlashInfer! Before we review your pull
request, please make sure the following items are complete.

### ✅ Pre-commit Checks

- [ ] I have installed `pre-commit` by running `pip install pre-commit`
(or used your preferred method).
- [ ] I have installed the hooks with `pre-commit install`.
- [ ] I have run the hooks manually with `pre-commit run --all-files`
and fixed any reported issues.

> If you are unsure about how to set up `pre-commit`, see [the
pre-commit documentation](https://pre-commit.com/).

## 🧪 Tests

- [ ] Tests have been added or updated as needed.
- [ ] All tests are passing (`unittest`, etc.).

## Reviewer Notes

<!-- Optional: anything you'd like reviewers to focus on, concerns, etc.
-->

Author: nvjullin

include/flashinfer/comm/trtllm_allreduce_fusion.cuh

@@ -1364,12 +1364,16 @@ cudaError_t allreduce_fusion_kernel_launcher(AllReduceFusionParams<T> const& par
     threads_per_block *= 2;
     cluster_size /= 2;
   }
+  int sm_count = get_sm_count();
+  while (cluster_num * cluster_size > sm_count && cluster_size > 1 && threads_per_block <= 512) {
+    threads_per_block *= 2;
+    cluster_size /= 2;
+  }
   FLASHINFER_CHECK(oneshot || threads_per_block >= params.nranks,
                    "not oneshot, or threads_per_block < nranks");
   int block_size = threads_per_block;
   FLASHINFER_CHECK(block_size <= 1024 && cluster_size > 0,
                    "block_size > 1024 or cluster_size <= 0");
-  int sm_count = get_sm_count();
   int grid_size = (std::min(sm_count, cluster_num * cluster_size) / cluster_size) * cluster_size;
   cudaLaunchConfig_t cfg;
   cudaLaunchAttribute attribute[2];