fix: preserve multi-value shape tensor inputs

[zupengwang]

Description

get_torch_tensor() assumed every shape tensor contained one value and always returned the first optimized shape value. Multi-value shape tensors were therefore converted to a scalar and could not be used for SymInt[] arguments such as aten.full's size.

This change preserves the existing scalar behavior for one-element shape tensors and returns the full optimized shape as a list for multi-element shape tensors. It also updates the related type annotations and adds a focused regression test.

No new dependencies are required.

Fixes #3202

Type of change

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

Validation

• Targeted Dynamo runtime, converter, and partitioning tests: 44 passed on each of two RTX 3090 GPUs
• Eager, torch.export, and TensorRT FP32/FP16 comparisons with static and dynamic shapes: maximum absolute error 0 on both GPUs
• Changed-file pre-commit hooks, mypy, Ruff, Black, py_compile, and git diff --check

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
• I have made corresponding changes to the documentation (not applicable: this does not change a user-facing API)
• 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 (left to maintainers and repository automation; no review was requested)

[zewenli98]

LGTM