Add generic RISC-V build path

[carlosqwqqwq]

Why

c-blosc reported riscv targets as unknown processors and left its generic fallback path noisy in shuffle.c, even though the generic C implementation was otherwise usable. This patch makes the baseline RISC-V build path explicit while preserving the existing generic implementation.

What changed

• Recognize riscv* targets explicitly in the top-level CMake processor selection instead of reporting them as unknown processors.
• Keep RISC-V builds on the existing generic shuffle/bitshuffle path without enabling SSE2 or AVX2 sources or flags.
• Let blosc/shuffle.c take a quiet generic fallback on RISC-V instead of emitting the non-x86 hardware-acceleration warning.
• Preserve the existing x86 acceleration logic and only adjust the conservative non-x86 build path.

Verification

• Ran native CMake configuration with Ninja on the local Windows host.
• Built the native tree successfully with cmake --build build-codex-native --parallel 4.
• Ran native test verification with ctest --test-dir build-codex-native --output-on-failure; all 1639/1639 tests passed.
• Ran simulated riscv64 CMake configuration with CMAKE_SYSTEM_NAME=Linux, CMAKE_SYSTEM_PROCESSOR=riscv64, CMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY, BUILD_SHARED=OFF, BUILD_STATIC=ON, BUILD_TESTS=OFF, BUILD_FUZZERS=OFF, BUILD_BENCHMARKS=OFF, and forced __riscv=1 / __riscv_xlen=64 macros because no real RISC-V toolchain is available on the host.
• Built the simulated riscv64 static library successfully with cmake --build build-codex-riscv --parallel 4.
• Checked build-codex-riscv/build.ninja: it contains shuffle-generic.c, bitshuffle-generic.c, shuffle.c, ZSTD_DISABLE_ASM, __riscv=1, and __riscv_xlen=64.
• Confirmed build-codex-riscv/build.ninja does not contain shuffle-sse2.c, bitshuffle-sse2.c, shuffle-avx2.c, bitshuffle-avx2.c, -msse2, -mavx2, or huf_decompress_amd64.S.

Notes

This is a conservative portability patch. It does not add RVV kernels or any RISC-V-specific performance optimizations. Verification covers native build and test stability plus simulated RISC-V target selection and source/flag exclusion checks, but not a real riscv64 cross-toolchain build or hardware run.

[kalvdans]

Remove the noisy compiler output for all non-accellerated architectures, as cmake has already warned about it.

[carlosqwqqwq]

Thank you for pointing this out. I removed the noisy generic fallback warning for all non-accelerated architectures and rechecked the native tests plus the simulated riscv64 build locally; I’m still learning RISC-V, so I really appreciate your guidance.