feat: add minimal returning calls for Add and Gemm

[voltjia]

Summary

• Add returning-style Call support for Add and Gemm only.
• Route returning calls through the shared Operator<Key>::Call(...) entry in src/operator.h, so future hooks can stay centralized.
• Let each base operator define only MakeReturnValue(...) for output construction; operator.h then reuses the existing in-place Call(..., out) path.
• Keep existing in-place Call overloads available via using Operator<Op>::Call.
• Add C++ API smoke coverage for returning Add::Call(...) and Gemm::Call(...), plus existing in-place API regression coverage in tests/test_cpp_api.py.

Motivation

Returning tensor APIs are easier for external users than explicitly allocating and passing output tensors. InfiniOps still does not own tensor storage: the returning path delegates allocation to the external tensor type through Empty(...), then calls the existing in-place operator implementation.

Closes N/A.

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

$ python -m pip install .[dev] --no-build-isolation \
    -C cmake.define.INFINI_RT_SOURCE_DIR=/workspace/InfiniRT \
    -C cmake.define.INFINI_OPS_SMOKE_BUILD=ON \
    -C cmake.define.AUTO_DETECT_BACKENDS=OFF \
    -C cmake.define.WITH_TORCH=OFF
Successfully built InfiniOps
Successfully installed InfiniOps-0.1.0

$ python -m pytest -v --tb=short --junitxml=/workspace/results/test-results.xml tests/test_cpp_api.py
collected 2 items

tests/test_cpp_api.py::test_cpp_operator_call_instantiation_smoke PASSED [ 50%]
tests/test_cpp_api.py::test_cpp_returning_call_smoke PASSED              [100%]

============================== 2 passed in 2.55s ===============================

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	smoke passed	Full suite not run; scoped API/header change.
Iluvatar	No	N/A - not affected	Not tested.
MetaX	No	N/A - not affected	Not tested.
Cambricon	No	N/A - not affected	Not tested.
Moore	No	N/A - not affected	Not tested.
Ascend	No	N/A - not affected	Not tested.

Full `pytest` output (optional)

See smoke output above.

Benchmark / Performance Impact

N/A. This adds a convenience API that allocates an output through the caller-provided tensor type and then reuses the existing in-place operator path.

Notes for Reviewers

• Depends on feat: add tensor-like TensorView constructor InfiniRT#10 for TensorView construction from tensor-like objects.
• tests/test_cpp_api.py compiles an external temporary C++ source against installed headers and links -linfiniops -linfinirt, so the new API is validated in the same form an external library would use it.
• First version intentionally supports only Add::Call(x, y) and Gemm::Call(a, b); other operators and more complex Gemm attribute forms are left out of scope.

[Ziminli]

这个是哪里需要呢？