Keyboard recording and display

[bytemines]

Summary

Keyboard keystroke recording and display for the video editor. Extracted from #1615 (keyboard-only — no captions timeline changes).

What it does:

• Captures keystrokes during screen recording (key down/up with timestamps)
• Groups rapid typing into display segments with configurable thresholds
• Renders keyboard overlays via WGPU with fade/bounce animations
• Provides interactive timeline track (drag, resize, split, delete)
• Settings panel for font family, weight, size, colors, animation timing
• Auto-regenerates segments when grouping settings change

Changes from #1615

Extracted keyboard-only code and applied fixes during extraction:

Correctness

• Fixed O(n²) modifier tracking → O(n) sliding window
• Shift no longer breaks segments — capitalization stays in same group
• Added CapsLock toggle tracking with Shift XOR logic
• Added Shift+symbol mapping (1→!, 2→@, etc. — full US QWERTY)
• Fixed RMeta (right Command) not recognized as modifier
• Fixed Space key mismatch (" " vs "Space") — special-key logic now works
• Backspace works even when show_special_keys is disabled
• Fixed stale current_keys in modifier group transitions
• Events sorted defensively inside group_key_events
• Fixed seconds/ms unit mismatch in display duration (was passed as 0.8s to Rust expecting ms, making all segments minimal length)
• Fixed split logic unit mismatch (timeOffsetMs compared against seconds)
• Uppercase keys in Shift combos (⌘+⇧+L not ⌘+⇧+l)
• Hotkey combos use + separator (⌘+⇧+L instead of ⌘⇧L)

Data integrity

• SingleSegment recordings now load keyboard data (was silently empty)
• Crash recovery restores keyboard.json (was hardcoded to None)
• Split segments properly partition displayText and keys at split point
• Override fields use skip_serializing_if to reduce JSON size
• Segment IDs use crypto.randomUUID() instead of Date.now()

Code quality

• Fixed Clippy violations (collapsible_if, unnecessary_lazy_evaluations, needless_borrow, unnecessary_cast)
• FromStr for KeyboardPosition returns Err for unknowns + #[derive(Default)]
• Extracted rendering magic numbers to named constants
• Removed redundant set_scissor_rect GPU call
• eprintln! → tracing::warn!
• Unit-suffixed field names throughout: fadeDurationSecs, displayDurationSecs, timeOffsetMs, groupingThresholdMs
• Renamed "linger duration" → "display duration" for clarity
• No code comments (project convention)

Frontend (SolidJS)

• Fixed getSetting type safety (removed Record<string, unknown> cast)
• selectedSegment wrapped in createMemo (proper reactivity)
• Added toast notifications on segment generation success/failure
• Extracted shared MIN_KEYBOARD_SEGMENT_SECS constant
• Font family dropdown (Monospace default, Sans-Serif, Serif)
• Font weight dropdown (Normal, Medium, Bold)
• Auto-regenerate segments when grouping/display settings change
• Fixed missing Input import in ConfigSidebar (crash when clicking segment)
• Semantic icons per settings field

Scope

• Zero captions code — fully independent from captions timeline migration

Architecture

Recording → cursor.rs polls device_query every ~16ms
         → key events flushed to keyboard.json per segment

Editor   → generate_keyboard_segments groups events by threshold
         → KeyboardTrack provides drag/resize/split/delete
         → KeyboardTab configures appearance and behavior

Render   → KeyboardLayer (WGPU) renders text overlay with fade/bounce
         → Included in both editor preview and MP4/GIF export

Follow-up

Mechanical keyboard sounds + click sounds will be a separate PR, using these keystroke timestamps and existing CursorClickEvent data mixed into the audio pipeline.

Test plan

• cargo test -p cap-project — 17 tests pass
• cargo check -p cap-desktop — full build passes
• cargo clippy — clean
• pnpm typecheck — only pre-existing web app errors
• Unit audit: all time values verified (seconds vs ms) across full pipeline
• Manual: record with keyboard enabled, verify overlay in editor
• Manual: export MP4, verify keyboard overlay in output
• Manual: split/delete/drag keyboard segments in timeline
• Manual: test Shift+symbols and CapsLock display
• Manual: change grouping/display settings and verify auto-regeneration