[AutoDeploy][1/n] Modular InferenceOptimizer

[lucaslie]

The interface and initial transforms in the new modular InferenceOptimizer design

[Copilot]

Pull Request Overview

Introduces a modular inference optimizer pipeline by defining a high-level InferenceOptimizer class and migrating existing export and cleanup logic into discrete, configurable transforms.

• Added a filtered_nodes utility to simplify node filtering by op type.
• Replaced the legacy export flow in transform.py with the new ModularInferenceOptimizer.
• Extracted input-constraint cleanup into its own transform and cleaned up the export code accordingly.

Reviewed Changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tensorrt_llm/_torch/auto_deploy/utils/node_utils.py	Added OpOrOverload alias and new filtered_nodes utility.
tensorrt_llm/_torch/auto_deploy/transformations/transform.py	Switched to ModularInferenceOptimizer, removed legacy export path.
tensorrt_llm/_torch/auto_deploy/transformations/export.py	Removed obsolete functions and relocated aliased-param load hook.
tensorrt_llm/_torch/auto_deploy/transform/optimizer.py	Introduced InferenceOptimizer entrypoint and transform loop.
tensorrt_llm/_torch/auto_deploy/transform/library/*.py	Added new wrapper transforms (export_to_gm, build_model, cleanup_input_constraints).
tensorrt_llm/_torch/auto_deploy/llm_args.py	Added transforms config for the new optimizer.
tensorrt_llm/_torch/auto_deploy/config/default.yaml	Populated default transform configurations.

Comments suppressed due to low confidence (1)

tensorrt_llm/_torch/auto_deploy/utils/node_utils.py:219

• [nitpick] Consider adding unit tests for filtered_nodes to validate that it correctly yields nodes matching single and multiple ops, and skips non-matching ones.

def filtered_nodes(