fix: export active implementation query

[voltjia]

Summary

• Generate extern template declarations and matching explicit instantiations for Operator<Op>::active_implementation_indices(Device::Type) in the installed operator call surface.
• Extend generator coverage and the installed C++ Add smoke so the query must link successfully and report native implementation index 0 on CPU.

Motivation

Downstream C++ consumers currently instantiate the query from installed public headers, where native implementation specializations are not visible, so a supported backend can incorrectly report no active implementations. Generating the query instantiation beside the existing operator call and default-implementation instantiations gives it the same specialization visibility as library dispatch.

No issue is linked.

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

No platform kernel or Python binding changes are included. The focused shared-codegen tests and the installed-prefix C++ API smoke passed.

$ python3 -m pytest -q tests/test_generate_wrappers.py
26 passed in 0.25s

$ python3 -m pytest -q tests/test_cpp_api.py::test_cpp_operator_call_instantiation_smoke
1 passed in 1.05s

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	No	N/A - No backend code changed.	N/A - No backend code changed.
Iluvatar	No	N/A - No backend code changed.	N/A - No backend code changed.
MetaX	No	N/A - No backend code changed.	N/A - No backend code changed.
Cambricon	No	N/A - No backend code changed.	N/A - No backend code changed.
Moore	No	N/A - No backend code changed.	N/A - No backend code changed.
Ascend	No	N/A - No backend code changed.	N/A - No backend code changed.

Benchmark / Performance Impact

N/A. The change only exports an existing query through generated C++ template instantiations.

Notes for Reviewers

Please focus on whether the query declaration and definition are generated in the same operator-specific translation units as the existing default-implementation lookup. The CPU build used WITH_TORCH=OFF, GENERATE_OPERATOR_CALL_INSTANTIATIONS=ON, and INFINI_OPS_OPS=add, then compiled and ran the smoke against the installed headers and libraries.