feat: expose returning add python call

[voltjia]

Summary

• Add opt-in Python returning overload generation for operators with MakeReturnValue, enabled only for add in this PR.
• Add Pybind11Tensor helpers for constructing returned PyTorch tensors through the external tensor protocol.
• Reuse the existing tests/test_add.py::test_add coverage: contiguous output cases call the returning overload, while existing in-place cases remain covered.

Motivation

Returning-style Python calls are more ergonomic for Python users, but InfiniOps still should not own tensor storage. This PR exposes that behavior through the existing external tensor construction protocol and keeps the rollout narrow to add.

Closes N/A - no linked issue.

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 (requires a ! in the Conventional Commits prefix or a BREAKING CHANGE: footer)

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

$ INFINI_RT_ROOT=/infinirt-prefix scripts/dev/build.sh nvidia --smoke --jobs 16 --torch-jobs 2 --binding-jobs 2
[build] platform  : nvidia
-- Wrapper op allowlist: add;mul;cast;cat;gemm;matmul;linear;rms_norm;swiglu;causal_softmax;abs;clamp;exp
warning: operator(s) not available for active devices: matmul
[build] done

$ CUDA_VISIBLE_DEVICES=1 PYTHONPATH=/workspace/build-nvidia/install:${PYTHONPATH:-} python3 -m pytest -q tests/test_generate_wrappers.py tests/test_add.py --devices nvidia
191 passed, 36 skipped in 7.00s

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	smoke build passed	Targeted tests/test_generate_wrappers.py tests/test_add.py --devices nvidia: 191 passed, 36 skipped
Iluvatar	No	N/A - not affected	N/A
MetaX	No	N/A - not affected	N/A
Cambricon	No	N/A - not affected	N/A
Moore	No	N/A - not affected	N/A
Ascend	No	N/A - not affected	N/A

Additional checks

ruff format --check scripts/generate_wrappers.py tests/test_add.py tests/test_generate_wrappers.py
3 files already formatted

ruff check scripts/generate_wrappers.py tests/test_add.py tests/test_generate_wrappers.py
All checks passed!

git diff --check origin/master
# passed

Benchmark / Performance Impact

N/A - API binding change only.

Notes for Reviewers

• matmul returning Python exposure is intentionally out of scope for this PR.
• Python returning overloads remain explicitly allowlisted through _PY_RETURNING_OPS; currently only add is enabled.
• The matmul smoke-build warning is expected on NVIDIA because this PR does not add a NVIDIA or PyTorch fallback implementation for matmul.