feat: add runtime backend API foundation

[voltjia]

Summary

• Add the non-graph runtime API surface introduced by CPU runtime extraction to the remaining runtime backends.
• Wire CUDA-compatible backends to their corresponding runtime APIs for host allocation, stream-ordered allocation/free where supported, mem info, memset async, stream, and event operations.
• Add conservative unsupported stubs for Ascend/Cambricon APIs without confirmed direct equivalents, and for Moore async allocation/free on the current Moore SDK.
• Extend the platform-adaptive native runtime and dispatch tests so the new API surface is exercised on each enabled backend.

Motivation

This splits the backend runtime API foundation out of #9 so that #9 can focus on graph runtime APIs only. It follows the surface already added for CPU in #8 and now uses the shared platform-adaptive test mechanism from #15.

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

ssh nvidia
image: accelerator-dev/nvidia:latest
cmake -B build -DWITH_CPU=ON -DWITH_NVIDIA=ON -DINFINI_RT_BUILD_TESTING=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build -j$(nproc)
ctest --output-on-failure
100% tests passed, 0 tests failed out of 7

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	yes	passed	CPU+NVIDIA CMake/CTest passed on ssh nvidia with accelerator-dev/nvidia:latest
Iluvatar	yes	passed	CPU+Iluvatar CMake/CTest passed on ssh iluvatar with accelerator-dev/iluvatar:latest
MetaX	yes	passed	CPU+MetaX CMake/CTest passed on ssh metax with accelerator-dev/metax:latest; installed CMake in-container for ctest
Cambricon	yes	passed	CPU+Cambricon CMake/CTest passed on ssh cambricon with accelerator-dev/cambricon:latest
Moore	yes	passed	CPU+Moore CMake/CTest passed on ssh moore with accelerator-dev/moore:latest; async alloc/free are expected unsupported on current SDK
Ascend	yes	passed	CPU+Ascend CMake/CTest output passed on ssh ascend with accelerator-dev/ascend:latest; SSH wrapper returned non-zero due known host issue

Representative NVIDIA `ctest` output

Test project /workspace/InfiniRT/build
Start 1: test_smoke
1/7 Test #1: test_smoke ....................... Passed 0.00 sec
Start 2: test_core
2/7 Test #2: test_core ........................ Passed 0.01 sec
Start 3: test_cpu_runtime
3/7 Test #3: test_cpu_runtime ................. Passed 0.00 sec
Start 4: test_nvidia_runtime
4/7 Test #4: test_nvidia_runtime .............. Passed 0.53 sec
Start 5: test_runtime_dispatch
5/7 Test #5: test_runtime_dispatch ............ Passed 0.47 sec
Start 6: test_install
6/7 Test #6: test_install ..................... Passed 0.01 sec
Start 7: test_install_consumer
7/7 Test #7: test_install_consumer ............ Passed 1.52 sec
100% tests passed, 0 tests failed out of 7

Benchmark / Performance Impact

N/A

Notes for Reviewers

• This PR intentionally does not add graph APIs; those remain in feat: add graph runtime API #9.
• Ascend/Cambricon get the same public surface but return non-success for APIs without a confirmed direct runtime equivalent, rather than pretending unsupported async/event operations succeeded.
• Moore's current SDK does not declare musaFreeAsync, so this PR exposes the API but reports async allocation/free as unsupported for Moore.