feat: add aten.glu converter

[zupengwang]

Description

Adds Dynamo converter support for torch.ops.aten.glu.default. The converter implements GLU as an equal split followed by sigmoid and elementwise multiplication, and conservatively requires the split dimension to be statically known and even while allowing other dimensions to remain dynamic.

This avoids a PyTorch fallback and extra TensorRT engine partitions for models that use GLU, including the Conformer use case described in #2197. No additional dependencies are required.

Validation performed locally:

• Converter tests cover FP32, FP16, dim=0/1/-1, rank 2/3, and a dynamic batch range.
• GLU, split, and sigmoid converter tests: 17 passed.
• Full-compilation torch.export checks on two RTX 3090 GPUs produced a single TensorRT module with no PyTorch fallback for FP32, FP16, and dynamic shapes; outputs matched eager PyTorch exactly in the tested cases.
• Black, Ruff, isort, mypy, and git diff --check passed for the changed files.

The full repository test suite and GitHub CI have not been run. The dynamic GLU split dimension remains unsupported; invalid odd split dimensions retain PyTorch's existing error behavior.

Fixes #2663

Type of change

• New feature (non-breaking change which adds functionality)

Checklist:

• My code follows the style guidelines of this project (You can use the linters)
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas and hacks (no non-obvious hacks were added)
• I have made corresponding changes to the documentation (not required; supported-op documentation is generated from the converter registry)
• I have added tests to verify my fix or my feature
• New and existing unit tests pass locally with my changes
• I have added the relevant labels to my PR in so that relevant reviewers are notified