Move LB8a and LB9 out of the table

[eggrobin]

Hopefully no functional change.

Last time I attempted to look at Unicode 15.1 line breaking, that was made impractical by the need, for every new state X, to add an X_ZWJ state, transitions X CM → X, X ZWJ → X_ZWJ, X_ZWJ CM → X, as well as X_ZWJ Y → Z for every transition X Y → Z, and to add or update rules to prevent breaks after X_ZWJ.

Hopefully this will make that upgrade a little more tractable. (Incidentally it makes the state table a bit smaller.)

Tested with 200 000 monkeys (recall that only 200 are checked in).

Related to #3255; see my comment there for the rationale.

Aside: While looking at this, it came to my attention that the LineBreakStrictness::Anywhere option does not do what the standard says, cf. https://drafts.csswg.org/css-text-3/#valdef-line-break-anywhere and https://drafts.csswg.org/css-text-3/#typographic-character-unit referenced therein. Of course, we do have a correct implementation of line-break: anywhere, since we have a grapheme cluster segmenter.

[robertbastian]

Can't tell for sure that this is correct, but if the tests pass I'm very happy that it removes 1k lines from the state table file, and 6k bytes from the data.