macOS: fix CJK punctuation dropped when insertText: called without hasMarkedText()

[zhjie]

Summary

The macOS Chinese Simplified Pinyin IME (and likely other CJK input methods) converts punctuation keys by calling insertText:replacementRange: without a prior setMarkedText: call. Because hasMarkedText() is false at that point, the previous code silently dropped the converted character and the raw ASCII key from NSEvent.characters reached the application.

Example: pressing , in Chinese input mode should produce ， (Chinese comma), but the application received , (ASCII) because ， was never emitted as Ime::Commit.

This affects at least:

• Chinese Simplified Pinyin (confirmed): all punctuation keys — ， 。 ？ ！ ； ： 「 」 etc.
• Likely other CJK IMEs that use direct insertText: for non-composition keys.

Root Cause

// Before: Commit only if we have marked text.
if self.hasMarkedText() && self.is_ime_enabled() && !is_control {
    ...emit Commit...
}
// Otherwise: silently dropped!

When an IME converts a key directly (no preedit), hasMarkedText() is false, so the committed text was dropped. The keyDown: callback then generated KeyboardInput with the raw NSEvent.characters (ASCII).

Fix

Emit Ime::Commit whenever is_ime_enabled() is true, regardless of hasMarkedText():

if self.is_ime_enabled() && !is_control {
    if self.hasMarkedText() {
        // Preserve existing preedit-clear-before-commit behaviour.
        self.queue_event(Ime::Preedit(String::new(), None));
    }
    self.queue_event(Ime::Commit(string));
    self.ivars().ime_state.set(ImeState::Committed);
}

ImeState::Committed is set in both paths, so keyDown: treats the key as handled (had_ime_input = true) and does not generate a duplicate KeyboardInput event with the raw ASCII character.

Testing

• Chinese Simplified Pinyin: punctuation (， 。 ？ ！ ； ：), full-width ASCII (ｆｏｒ), and composed characters (nihao → 你好) all work correctly.
• No duplicate ASCII character emitted alongside Ime::Commit.
• Standard preedit-then-commit sequence (setMarkedText → insertText) unchanged.
• Japanese IME — needs testing (architecture similar to Chinese, expected no regression)
• Korean IME — needs testing (see also macOS: Fix Korean composing text commit key handling #4478 which touches related code)

Relation to #4478

PR #4478 fixes Korean-specific double-input and key-loss bugs. This PR is orthogonal: it fixes the case where insertText: is called without any prior setMarkedText:, which the Korean PR does not address.