fix: support dynamic slice_scatter fallback shapes

[kiwigitops]

Description

The slice_scatter fallback now expands its index tensor inside TensorRT when the source has dynamic dimensions. Static sources continue to use the existing NumPy broadcast path.

The fallback previously passed src.shape directly to np.broadcast_to. TensorRT represents dynamic dimensions as -1, which caused NumPy to raise before the network could be built.

A converter regression test covers a stepped slice_scatter with dynamic batch and feature dimensions.

Fixes #4437

Type of change

• Bug fix (non-breaking change which fixes an issue)

Validation

• Repository pre-commit hooks on both changed files (including isort, mypy, ruff, black, and typos)
• Python bytecode compilation for both changed files
• CUDA/TensorRT converter test was not run locally because this machine does not have CUDA or TensorRT

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have added a regression test for the fix