[Feature]: AutoDeploy: develop generalized pattern matcher for bmm-style MoE

🚀 The feature, motivation and pitch

#9556 introduces a graph-traversal style pattern matcher that is hard to maintain and currently only tested / hard-coded to Llama4.

Let's develop a pattern matcher in the style of the attention pattern matcher (see https://github.com/NVIDIA/TensorRT-LLM/blob/main/tensorrt_llm/_torch/auto_deploy/transform/library/attention.py) that enables flexible matching for multiple patterns and is more easily maintainable and extendable.

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.