fix(ops): copy convolution tensor metadata views

[voltjia]

Summary

• Copy convolution tensor shapes and strides into owned metadata containers.
• Restore compatibility with the current InfiniRT TensorView metadata-view API.
• Keep all convolution operator names, parameters, outputs, and execution behavior unchanged.

Motivation

InfiniRT now returns ShapeView and StridesView from lvalue TensorView objects. The direct assignments introduced with the canonical convolution operators in #882 therefore fail to compile against current InfiniRT. Explicit Tensor::Shape{...} and Tensor::Strides{...} construction matches the existing bias metadata path and works for both owned containers and metadata views.

Type of Change

• feat - new feature / new operator / new platform
• fix - bug fix
• perf - performance improvement (no behavioral change)
• refactor - code restructuring without behavior change
• test - adding or fixing tests only
• docs - documentation only
• build / ci - build system or CI configuration
• chore - tooling, formatting, or other non-code changes
• Breaking change

Platforms Affected

• CPU (WITH_CPU)
• NVIDIA (WITH_NVIDIA)
• Iluvatar (WITH_ILUVATAR)
• MetaX (WITH_METAX)
• Cambricon (WITH_CAMBRICON)
• Moore (WITH_MOORE)
• Ascend (WITH_ASCEND)
• PyTorch C++ bindings (WITH_TORCH)
• Build system / CMake / CI
• Python bindings / user-facing API

Smoke Test Result

scripts/dev/build.sh nvidia --jobs 16
[build] done
PYTHONPATH=build-nvidia/install pytest -q tests/test_conv2d.py --devices nvidia --maxfail=1
18 passed, 1 warning in 7.40s

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
CPU	Yes	wrapper build passed with WITH_CPU=ON	No CPU convolution provider
NVIDIA	Yes	build passed	focused conv2d: 18 passed
Iluvatar	Yes	not run locally	pending platform CI
MetaX	Yes	not run locally	pending platform CI
Cambricon	No	N/A	N/A
Moore	Yes	not run locally	pending platform CI
Ascend	No	N/A	N/A

Benchmark / Performance Impact

N/A. This only makes existing metadata copies explicit.

Notes for Reviewers

• No public operator schema changes are included, so the PyTorch alignment established in feat(ops): add PyTorch-aligned convolution operators #882 is unchanged.
• clang-format 21.1.8 --dry-run --Werror src/common/op_utils/conv.h passed.
• The dependent InfiniCore adapters for Conv2d, Copy, Relu, and SiluAndMul compile successfully against this commit and current InfiniRT.