To address the regression in zh stroke collation since CLDR 30, temporarily revert to the CLDR 29 stroke collation for CLDR 32. The real fix is to fix the tooling, for which we have : but that cannot be done in CLDR 32.
Revert stroke collation full & short to data as of r11973, before some basic chars were deleted in r12662
Whatever... This takes the zh stroke order back to Unicode 8.