feat: compact single-panel widget with logo bolt

[TimInTech]

Summary

The main window is now a single compact, frameless dark surface instead of a native window with separate floating panels. The PR now also carries the documentation, screenshots and settings fixes that go with it.

• One surface: title, minimize/close, gold microphone with recording waves and processing ring, status/timer and the tool row (editor, history, TTS, settings, overflow menu). Workflows, collect dictation and discard stay in the transient popup.
• Window chrome: frameless; the window moves via QWindow.startSystemMove() (works on Wayland). The close button hides to the tray, minimize minimizes the real window, and the tray restores a minimized window with showNormal().
• Compact size: 242 × 213 px on GNOME/Wayland (previously 288 × 300 px with two floating surfaces).
• Branding: the official Blitztext bolt is read locally from app/assets/logo-mark-dark.svg (identical to the website brand asset) and cut out of the microphone capsule as a vector path. No network access at runtime.
• Crash fix: a PyQt6 QPoint/QPointF TypeError aborted RecordButton.paintEvent.
• Neumorphic restyling of the compose, history and TTS windows.

Settings and i18n fixes (9213108)

• The API key status word was hardcoded German, so the English settings tab read "Status: nicht gesetzt".
• Emoji density offered the raw German config values in both languages; the caption is translated now while the stored value stays wenig/mittel/viel.
• Qt swallowed the & in captions such as "Insert & Close", "Text & AI" and "Dictate & improve". The new i18n.tq() escapes it for tabs, buttons and menu actions.
• Field help now reaches editors that sit inside layout rows (base URL), so keyboard and screen reader users keep the HTTPS guidance. This addresses the open review comment on app/blitztext_linux.py.

Documentation and assets (0de779a)

• All screenshots are generated into docs/screenshots/linux/ again; the legacy aliases (main-window*, settings-ai-workflows*, Banner.png) are gone and there is exactly one current set.
• New views: idle / recording / processing / options menu, the compose prompt editor, and a regenerated banner, social preview and demo GIFs.
• README.md and README.de.md document the widget, the five text actions (instead of the removed nine writing-style presets), the current workflow names and the "Text & AI" settings tab.
• Settings screenshots are sized per tab, and the read-aloud shot shows Piper instead of a cloud error.
• scratch/ and prototypes/ are ignored locally.

Test plan

• WHISPER_GUI_TESTS=1 QT_QPA_PLATFORM=offscreen pytest: 684 passed
• QT_QPA_PLATFORM=offscreen pytest: 547 passed, 137 skipped
• git diff --check and py_compile clean
• Real GNOME/Wayland state cycle (idle → recording → transcribing → AI processing → idle) without paint errors, stable window size
• App started via ./run.sh without errors
• Every image referenced by both READMEs exists; no orphaned screenshots left
• Manual check on GNOME: drag, minimize, restore from tray

Related

The website repository TimInTech/blitztextweb was updated in lockstep (same screenshots, same wording) in 1428b8a.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QpE6yNpSWEnYfvJd4hyJK3

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: eff35f831c

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".