Respect riscv targets in platform macros

[carlosqwqqwq]

Why

RapidJSON is header-only, but some platform macros and test/performance entry points still infer target capabilities from host compiler predefined macros. In a simulated or cross-target riscv64 configuration built from an x86_64 host compiler, that can incorrectly preserve x86-only assumptions such as the 48-bit pointer optimization or host SIMD auto-detection. This patch makes the generic RISC-V path explicit and keeps simulated validation builds from inheriting host-specific SIMD behavior.

What changed

• Add a conservative RAPIDJSON_RISCV platform macro in rapidjson.h.
• Treat __riscv_xlen == 64 as a 64-bit target in rapidjson.h.
• Disable RAPIDJSON_48BITPOINTER_OPTIMIZATION on RISC-V so RISC-V builds do not inherit the x86-64-specific pointer compression assumption.
• Update test/perftest/perftest.h and test/unittest/simdtest.cpp so simulated or cross-target RISC-V validation does not auto-enable host SSE2/SSE4.2/NEON paths.
• Adjust the tutorial example to use a dynamically built std::string, avoiding an x86-specific short-string layout assumption during non-x86 validation builds.

Verification

• Ran native CMake configuration with Ninja using RAPIDJSON_BUILD_DOC=OFF and RAPIDJSON_BUILD_TESTS=OFF.
• Built the native examples successfully with cmake --build build-codex-native-058 --parallel 4.
• Ran simulated riscv64 CMake configuration with CMAKE_SYSTEM_NAME=Linux, CMAKE_SYSTEM_PROCESSOR=riscv64, CMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY, and forced __riscv=1 / __riscv_xlen=64 because no real RISC-V toolchain is available on the host.
• Preprocessed rapidjson.h under forced __riscv=1 / __riscv_xlen=64 and confirmed RAPIDJSON_RISCV=1, RAPIDJSON_64BIT=1, and RAPIDJSON_48BITPOINTER_OPTIMIZATION=0.
• Built the simulated example sources to object files, including example/tutorial/tutorial.cpp.o, and confirmed build.ninja contains no -msse*, -mavx*, or NEON flags.

Notes

This is a conservative portability patch. It does not add a RISC-V SIMD backend. Because the host environment uses a Windows-native compiler while simulating a Linux riscv64 target, the final executable link step still inherits a host/toolchain mismatch around -rdynamic. Verification therefore focuses on native example build stability, simulated RISC-V configuration, object compilation, macro checks, and x86/ARM SIMD flag exclusion rather than a full real riscv64 link.

[tencent-adm]

All committers have signed the CLA.