Update vendored zstd to 1.5.6

[Viq111]

Same workflow as #125

• Run below:

git clone https://github.com/facebook/zstd.git c
cd c
git checkout v1.5.6
cd ..
git clone [email protected]:DataDog/zstd.git go
find c/lib \( -name '*.h' -o -name '*.c' -o -name '*.S' \) -exec cp -v {} go/ \;

• Run python tools/flatten_imports.py
• Run python tools/insert_libzstd_ifdefs.py

This is the code as of https://github.com/facebook/zstd/releases/tag/v1.5.6 (facebook/zstd@794ea1b)

Benchmarks will come in the next comment

[Viq111]

Benchmarks on a Apple M1 Max:

goos: darwin
goarch: arm64
pkg: github.com/DataDog/zstd
                       │     old     │                new                 │
                       │   sec/op    │   sec/op     vs base               │
BulkCompress-10          900.1n ± 2%   926.6n ± 1%  +2.94% (p=0.000 n=10)
BulkDecompress-10        473.8n ± 0%   484.3n ± 0%  +2.23% (p=0.001 n=10)
CtxCompression-10        66.77m ± 0%   67.31m ± 1%  +0.81% (p=0.004 n=10)
CtxDecompression-10      9.392m ± 0%   8.633m ± 0%  -8.08% (p=0.000 n=10)
StreamCompression-10     74.13m ± 2%   74.50m ± 1%       ~ (p=0.143 n=10)
StreamDecompression-10   11.17m ± 1%   10.28m ± 1%  -7.95% (p=0.000 n=10)
Compression-10           67.07m ± 1%   67.73m ± 1%  +0.99% (p=0.001 n=10)
Decompression-10         9.422m ± 1%   8.701m ± 1%  -7.66% (p=0.000 n=10)
geomean                  1.855m        1.815m       -2.14%

                       │      old      │                 new                  │
                       │      B/s      │      B/s       vs base               │
BulkCompress-10           97.47Mi ± 2%    94.69Mi ± 1%  -2.86% (p=0.000 n=10)
BulkDecompress-10         114.7Mi ± 0%    112.2Mi ± 0%  -2.18% (p=0.001 n=10)
CtxCompression-10         142.4Mi ± 0%    141.3Mi ± 1%  -0.80% (p=0.004 n=10)
CtxDecompression-10      1012.4Mi ± 0%   1101.4Mi ± 0%  +8.80% (p=0.000 n=10)
StreamCompression-10      128.3Mi ± 2%    127.6Mi ± 1%       ~ (p=0.138 n=10)
StreamDecompression-10    851.2Mi ± 1%    924.8Mi ± 1%  +8.64% (p=0.000 n=10)
Compression-10            141.8Mi ± 1%    140.4Mi ± 1%  -0.98% (p=0.001 n=10)
Decompression-10         1009.2Mi ± 1%   1092.9Mi ± 1%  +8.29% (p=0.000 n=10)
geomean                   266.2Mi         272.0Mi       +2.18%

Overall the performance seems to be very similar. Bulk* APIs seems to have taken a 2-3% hit, Decompression speed got a nice 7-8% boost.

We will cut a pre-release and test in our environments