fix: align graph runtime API with runtime namespace

[voltjia]

Summary

• Move generated C++ runtime wrappers under infini::rt::runtime so that API names matching CUDA Runtime API stay in that nested namespace.
• Keep the outer infini::rt namespace focused on runtime-device selection APIs.
• Adapt graph C++ tests and C API implementation to the runtime namespace; C API graph handles remain opaque wrappers around backend/raw handles.
• Generate graph runtime declarations only when an enabled backend implements graph APIs, so CPU-only builds do not expose unsupported graph symbols.
• Refresh docs and install-consumer tests for the new public API layout.

Motivation

This helper PR lets feat/graph-runtime-api absorb the current master runtime API layout before the graph-runtime PR is rebased. It preserves the project decision that CUDA-aligned Runtime API names live under infini::rt::runtime, while InfiniRT-specific selection APIs live directly under infini::rt.

No issue is closed directly by this PR.

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 --test-dir build --output-on-failure
100% tests passed, 0 tests failed out of 9

# CPU-only generated API check, same image
cmake -B build-cpu -DWITH_CPU=ON \
-DINFINI_RT_BUILD_TESTING=ON \
-DCMAKE_BUILD_TYPE=Release
cmake --build build-cpu -j8
ctest --test-dir build-cpu --output-on-failure
100% tests passed, 0 tests failed out of 6

clang-format 21 dry-run: passed
ruff format --check scripts/generate_public_headers.py: passed

# Integration check with simulated #13 merged into #9 and rebased onto master,
# InfiniCore#1351, InfiniLM commit 36e2207, model Qwen3-0.6B.
python3 examples/test_infer.py --device nvidia --enable-graph \
--model=/data-aisoft/mechdancer/models/Qwen3-0.6B/
Standalone InfiniRT graph bridge loaded: /workspace/verify/infini-rt-prefix/lib/libinfinirt.so
Using standalone InfiniRT graph bridge for graph capture and replay.
[codex-verify] InfiniRT C API infiniRtStreamBeginCapture: 1
[codex-verify] InfiniRT C API infiniRtStreamEndCapture: 1
[codex-verify] InfiniRT C API infiniRtGraphInstantiate: 1
[codex-verify] InfiniRT C API infiniRtGraphLaunch: 511
Inference completed, total_time: 1448.2 ms

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	smoke passed	Full graph integration passed with InfiniCore#1351 and InfiniLM as above.
Iluvatar	Yes	not rerun in this helper PR	API/generation touched; backend validation to run on the rebased #9 branch.
MetaX	Yes	not rerun in this helper PR	API/generation touched; backend validation to run on the rebased #9 branch.
Cambricon	Yes	not rerun in this helper PR	API/generation touched; backend validation to run on the rebased #9 branch.
Moore	Yes	not rerun in this helper PR	API/generation touched; backend validation to run on the rebased #9 branch.
Ascend	Yes	not rerun in this helper PR	API/generation touched; backend validation to run on the rebased #9 branch.

Full `pytest` output (optional)

N/A - no pytest suite in this PR.

Benchmark / Performance Impact

N/A - this PR changes API placement, wrapper generation, tests, and docs; it is not performance-sensitive.

Notes for Reviewers

• This is a helper PR targeting feat/graph-runtime-api, not master directly.
• It intentionally keeps CUDA-aligned names under infini::rt::runtime; outer infini::rt remains available for InfiniRT-specific device-type selection APIs.
• Current InfiniLM main has newer InfiniCore RoPE/mrope API expectations than InfiniCore#1351, so the graph integration smoke used InfiniLM commit 36e2207ae6eaa4296b4564895648d50aaeef5453, which is compatible with #1351.