Segmenter results are inconsistent across platforms

[sven-oly]

With data generated by ICU4J, I note these problems:

• ICU4X has 2 failures in Myanmar for word and line break (label 29 and 31)
• ICU4C shows a failure for Japanese in line break (label 11)

[sven-oly]

These should be reported to the ICU4X and ICU4C teams (maybe ICU4J, too).

Note also that test data should be generated from ICU or CLDR sources instead of the hand-crafted data now used.