Fix Unicode 15.0 sentence segmentation

[eggrobin]

Fix #4038.

Seems to work now; tested with 1 000 000 test cases (I’m not going to check that in) and the last failure I encountered so far was #‌79. I should try an overnight test though.

Outline of the monkey testing process:
In ICU, at https://github.com/eggrobin/icu/tree/export-monkeys-15.0:

# Note that you get one tenth of the loop count for sentence segmentation.
.\icu4c\source\test\intltest\x64\Release\intltest.exe "rbbi/RBBITest/TestMonkey@type=sent seed=1729 loop=100000 scalars_only export=meow"

In ICU4X:

cd components\segmenter
git checkout la-vache/main .\tests\testdata\SentenceBreakExtraTest.txt
Add-Content .\tests\testdata\SentenceBreakExtraTest.txt (Get-Content ..\..\..\icu\meow_sent.txt)
cargo test sentence_break_extra_test

Stare at the weeping tests (you will want to cherry-pick the traces from 21aca87), fix something in the state-transition table,

cd ..\..; cargo make bakeddata components/segmenter; cd components\segmenter

And back to

cargo test sentence_break_extra_test

Repeat until the heat death of the universe.

[robertbastian]

Looks good from what I can tell, which excludes the rules themselves