refactor(tuning): simplify configuration and persistence

[voltjia]

Summary

• Simplify Config auto-selection state and make operator calls resolve the device/configuration once before construction or dispatch.
• Replace the hand-written tuning JSON reader/writer with privately linked nlohmann/json, centralize environment initialization, and remove src/tuning_utils.h.
• Deduplicate generated module initialization and add focused generator plus C++ tuning-cache round-trip coverage.

Motivation

This is a stacked cleanup for InfiniTensor/InfiniOps#879, based on its head commit b2d3168d. The goal is to reduce branching, duplicated configuration handling, and custom serialization code without changing the feature's intended behavior.

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

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

Validated in accelerator-dev/nvidia:latest with InfiniRT 3fc917066e427351345037c9ef4ef56a5221b30a built with CPU and NVIDIA support.

INFINI_RT_ROOT=/infinirt-prefix \
INFINI_OPS_INSTALL_PREFIX=/workspace/build-nvidia/install/infini \
CPLUS_INCLUDE_PATH=/usr/local/cuda/include \
scripts/dev/test.sh nvidia --smoke

47 passed, 10 skipped, 6859 deselected
0 failed

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
CPU	Yes	passed as part of NVIDIA smoke build and C++ consumer tests	smoke only
NVIDIA	Yes	smoke passed	full suite not run
Iluvatar	Yes	not available locally	CI pending
Hygon	Yes	not available locally	no default PR job
MetaX	Yes	not available locally	CI pending
Cambricon	Yes	not available locally	CI pending
Moore	Yes	not available locally	CI pending
Ascend	Yes	not available locally	CI pending

Benchmark / Performance Impact

N/A. This PR is a code-structure cleanup and does not claim a performance change.

Notes for Reviewers

• Review this as a delta on top of feat(auto-tuing): add a new auto-tuning system InfiniTensor/InfiniOps#879; its base branch is feat/auto-tuning.
• nlohmann/json is a private dependency. CMake first reuses an installed 3.12.0 package, then falls back to the pinned 3.12.0 release archive and SHA-256.
• The C++ cache test covers environment initialization, JSON persistence, second-process lookup, and invalid-JSON cache miss behavior.