test: Enable xfailed trtllm decode long seqlen tests and update microbenchmark (#2018)

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

## 📌 Description


[tests/attention/test_trtllm_gen_attention.py](https://github.com/flashinfer-ai/flashinfer/blob/v0.5.0rc2/tests/attention/test_trtllm_gen_attention.py#L1021-L1076)
was failing and therefore marked xfail.

PR #2002 fixed the underlying root cause. Current PR thus removed the
`xfail` marker so that these long seqlen cases could be fixed moving
forward.

Additionally, PR #2002 revealed a bug in the microbenchmark script where
[trtllm_batch_decode_with_kv_cache](https://github.com/flashinfer-ai/flashinfer/blob/v0.5.0rc2/flashinfer/decode.py#L2082-L2083)
explicitly requires the workspace to be zeroed before first use:
```
    workspace_buffer : torch.Tensor. Must be initialized to 0 for its first use.
        workspace
```
while the microbenchmark code does not zero out, causing undefined
behavior such as IMAs that depend on the ordering of backends tested.
Current PR fixes the issue by explicitly calling
`workspace_buffer.zero_()` between testing different backends.


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

## 🔍 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

- [x] I have installed `pre-commit` by running `pip install pre-commit`
(or used your preferred method).
- [x] I have installed the hooks with `pre-commit install`.
- [x] 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.
-->


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Bug Fixes**
* Improved stability of performance benchmarks by properly resetting
workspace buffer between backend invocations.

* **Tests**
  * Enabled previously skipped test for long sequence length handling.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Author: bkryu

benchmarks/routines/attention.py

@@ -508,6 +508,8 @@ def run_backend_wrapper(backend):
     has_reference_output = False
     # Iterate over each backend:
     for cur_backend in backends:
+        # Clear workspace buffer to prevent unexpected interactions between backends.
+        workspace_buffer.zero_()
         if run_refcheck:
             outputs[cur_backend] = run_backend_wrapper(cur_backend).detach().clone()
             if cur_backend == "fa2":
@@ -975,6 +977,8 @@ def run_backend_wrapper(backend):
     has_reference_output = False
     # Iterate over each backend:
     for cur_backend in backends:
+        # Clear workspace buffer to prevent unexpected interactions between backends.
+        workspace_buffer.zero_()
         if run_refcheck:
             outputs[cur_backend] = run_backend_wrapper(cur_backend).detach().clone()
             if cur_backend == "fa2":
@@ -1427,6 +1431,8 @@ def run_backend_wrapper(backend):
     has_reference_output = False
     # Iterate over each backend:
     for cur_backend in backends:
+        # Clear workspace buffer to prevent unexpected interactions between backends.
+        workspace_buffer.zero_()
         if run_refcheck:
             outputs[cur_backend] = run_backend_wrapper(cur_backend).detach().clone()
             if cur_backend == "fa2":
@@ -1822,6 +1828,8 @@ def run_backend_wrapper(backend):
     has_reference_output = False
     # Iterate over each backend:
     for cur_backend in backends:
+        # Clear workspace buffer to prevent unexpected interactions between backends.
+        workspace_buffer.zero_()
         if run_refcheck:
             outputs[cur_backend] = run_backend_wrapper(cur_backend).detach().clone()
             if cur_backend == "fa2":

tests/attention/test_trtllm_gen_attention.py

@@ -1133,7 +1133,6 @@ def test_trtllm_batch_decode_long_sequence_length(
     head_dim,
 ):
     # Small number of test cases for long sequence length
-    pytest.xfail("trtllm-gen decode gets incorrect output with Long sequence length")
     _test_trtllm_batch_decode(
         "trtllm-gen",
         kv_layout,