Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>

…PdlOverlapWithNext;Remove DeepSeekV3 float32 logits constraint from
kernel launchers

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

## 📌 Description

1. Add dynamic block kernel (`routingIndicesDynBlockKernel`) comes from
the TensorRT-LLM. NVIDIA/TensorRT-LLM#12456 .
Made related modification by refactoring
 `LAUNCH_ROUTING_CUSTOM` with `dispatchRoutingPolicy` and `queryDispatchedMaxExperts`
2. Simplify PDL (Programmatic Dependent Launch) for routing kernels, as
the bug related to PDL is solved.
3. Added a default fallback tier (`Tier<1024, 32>`) to support future
models with >512 experts using the DeepSeek nGroup≤1 / MiniMax2 routing
policy.
4. Remove DeepSeekV3 float32 logits constraint
5. Improve policy tier dispatch error messages

## 🔍 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
```
python -m pytest tests/moe/test_trtllm_gen_fused_moe.py -k "test_dyn_block_kernel_routing or test_tier_1024_experts_routing" -xvs
python3 -m pytest tests/moe/test_trtllm_gen_fused_moe.py -k "test_routing_dtype_flexibility" -xvs
```


- [x] Tests have been added or updated as needed.
- [x] 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

* **New Features**
* Dynamic single-block routing for small token/expert workloads to
improve performance.

* **Improvements**
* Added a 1024-expert policy tier and clearer tier-dispatch diagnostics.
* More flexible routing/logits dtype handling and simplified kernel
overlap/launch synchronization.
* Automatic histogram-thread sizing and removal of legacy
cluster-size/overlap public flags.
* Autotuner now initializes packed TopK with per-token unique expert
IDs.

* **Tests**
* Added tests for dynamic-block routing, 1024-expert tier, MiniMax2
routing, and routing dtype combinations.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Christina Zhang <83400082+ChristinaZ@users.noreply.github.com>