RISC-V: gate __builtin_ctzll behind Zbb to avoid 10 % slowdown

[anthony-zy]

RISC-V: use __builtin_ctzll only when Zbb is present

On RV targets that implement the Zbb extension the builtin
expands to a single `ctz` instruction and is ~10 % faster.
Without Zbb the generic software sequence is emitted, which
actually slows us down by the same amount.  Gate the builtin
behind the predefined macro `__riscv_zbb` so we only use it
when the hardware supports it.

	默认启用CTZ不检查zbb	默认启用CTZ检查zbb
编译项不带zbb	42.5 MB/s	47.4 MB/s	环境不检查zbb，压缩下降10%
编译项带zbb	50.9 MB/s	50.9 MB/s	环境支持zbb，压缩性能提升19.7%

[danilak-G]

I think it's better to fix CMakeLists.txt and BUILD.bazel

[anthony-zy]

I think it's better to fix CMakeLists.txt and BUILD.bazel

Thanks for the suggestion. I've made the necessary changes to CMakeLists.txt .