expose complex-scripts feature (#24094)

[unclepomedev]

Objective

Fixes #24094

(Rebase of #24683, closed due to merge conflicts on main. Re-verified the fix still works on current main. That PR had received a community review approval.)

Text containing complex scripts that lack explicit word boundaries (CJK, Thai, Khmer, Lao, Myanmar) is segmented by parley's lightweight segmenter, which emits a data error per layout:

ICU4X data error: No segmentation model for language: ja

parley added a complex-scripts Cargo feature that switches to dictionary-based segmentation, but it was not reachable from Bevy.

Solution

• Add an opt-in complex_script_segmentation feature to bevy_text that forwards to parley/complex-scripts, propagated through bevy_internal and bevy. It is off by default because the bundled segmentation dictionary increases binary size (see below).

This supplies the segmentation data and enables dictionary-based word segmentation; it does not change CJK line-break positions, which ICU keeps permissive by design.

Testing

• cargo build -p bevy_text -p bevy_ui -p bevy_ui_widgets, with default features and with --features complex_script_segmentation.
• cargo tree -e features -i parley confirms parley's complex-scripts feature is only enabled when complex_script_segmentation is requested (not pulled in by any default/collection feature).
• Ran a Text2d example with a Japanese string in a width-bounded box (TextBounds, LineBreak::WordBoundary) under RUST_LOG=warn:
  • default: the No segmentation model for language: ja error is logged.
  • --features complex_script_segmentation: the error is gone.
• cargo run -p ci (lints + compile + tests). One unrelated local failure: bevy_ecs ... filtered_backtrace_test, which also fails on main on my machine (an environment-dependent backtrace-symbolization test, unrelated to this PR).
• Tested on macOS (Apple Silicon).

Binary size impact

Measured with a stripped release build of the text_debug example:

• without complex_script_segmentation: 57,018,696 bytes
• with complex_script_segmentation: 60,832,984 bytes
• delta: ~3.6 MiB / ~6.7%

It seems the feature should be an opt-in.