lo128.dat and hi128.dat were constructed manually, then the
256.bytes.two-concatenated-streams.xz file was constructed by:
xz --keep lo128.dat
xz --keep hi128.dat
cat lo128.dat.xz hi128.dat.xz > 256.bytes.two-concatenated-streams.xz
It was created wih a custom xz binary (built from the source at
https://github.com/tukaani-project/xz), with this command:
xz --compress --keep --stdout enwik5 > enwik5.lzma2-chunk-max-1-13.xz
after applying this patch to xz's src/liblzma/lzma/lzma2_encoder.h:
-#define LZMA2_CHUNK_MAX (UINT32_C(1) << 16)
+#define LZMA2_CHUNK_MAX (UINT32_C(1) << 13)
The resultant enwik5.lzma2-chunk-max-1-13.xz file has multiple LZMA2
chunks, unlike the existing enwik5.xz (created with the normal chunk
limit). Wuffs' std/xz currently does not decode this new file properly.