fix(terminal): avoid duplicate broadcast keypress text

[binaricat]

Summary

Typing uppercase letters and spaces with cross-tab broadcast enabled could send some characters twice to recipient terminals while the source looked correct. Preserve the physical-key pairing when xterm emits its text on keypress, and prevent a trailing insertText marker from treating the next physical key as a second composition commit.

Type of Change

• Bug fix
• New feature
• Refactor / code cleanup
• Documentation update
• Build / CI change
• Other

Related Issue (optional)

Found while smoke testing #3275 in the development app. No separate issue.

Changes Made

• Re-arm the existing legacy-data pairing on a matching keypress; retain the existing unmatched-event cleanup so later paste is not swallowed.
• Give an identified physical-key payload precedence over a stale composition fallback marker. Actual composition/input handlers clear physical pairing before delivering committed text.
• Add regression coverage that executes the actual registered key, input, and data callbacks with controlled event timing and the real recipient encoder, without constructing the unrelated renderer/addon stack.

Screenshots / Demo

Verified in the actual npm run dev Electron UI on macOS, with local zsh and a Linux SSH session:

• Before: echo NETCATTY_SMOKE_BROADCAST_20260906_A looks correct on the source, but recipients repeat uppercase characters. Individual Shift+A/B/C produces ABBC at the recipient.
• Captured event order: Space keypress emits data, followed by insertText; the next physical A arrives before the composition fallback timer expires and is broadcast again as text.
• After: full uppercase test strings match exactly in both directions; individual Shift+A/B/C produces exactly ABC.
• Disabling broadcast keeps a subsequent marker on the source only.
• Pasting echo "中文 ABC paste" still reaches the recipient correctly and produces the expected text once.

Temporary diagnostic recording was removed before final verification.

Testing

• Tested locally with npm run dev and Computer UI actions
• npm run lint passes; final changed files also pass focused lint
• npm test: 11,484 passed, 18 skipped, zero failures
• Generated capability tool specs: not applicable
• No new console errors or warnings — not asserted for all application flows

118 focused keyboard, broadcast, IME, and terminal tests pass. The baseline fails six delayed-keypress regression cases; the intermediate timing repair still fails the captured trailing-insertText case ( AABBCC instead of ABC). The final change passes all nine new cases, including keyless Chinese text and subsequent paste.

Checklist

• Existing project style and input architecture followed
• Behavior and validation documented above
• No intentional breaking changes

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-06T08:55:24.841221Z	f15508a	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[binaricat]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. More of your lovely PRs please.

Reviewed commit: f15508aaff

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[netcatty-bot]

Codex reported no major issues. This PR is marked ready for human review/merge.