Zoom rect now at text area left edge (WINDOW_TEXT_TO_FRAME_PIXEL_X)
with cursor-width (FRAME_COLUMN_WIDTH) instead of full window width.
Face matching adds 'selection' (company-tooltip-selection).
Zoom now tracks the selected candidate row via overlayZoomRect
instead of always pointing at the text cursor. Returns to cursor
on typing (chars_modiff change).
Previous two-reference algorithm failed because:
- Vertico's cursor-space line (face=nil) confused the reference
- Count overlay processed before candidates overlay
- Group titles have distinct faces too
New approach: ns_ax_face_is_selected checks if face symbol name
contains 'current' or 'selected'. Works for all major frameworks
(Vertico, Icomplete, Ivy) without framework-specific code.
Key changes from previous version:
- Remove overlay text from ns_ax_buffer_text (was causing spurious
'new line' announcements via VoiceOver text diff)
- Do NOT invalidate text cache on overlay change
- Two-reference face detection (handles selected at any position)
- SDATA scan instead of per-char Faref for newline detection
- Zoom tracking via UAZoomChangeFocus for selected candidate row
- Deduplication via cachedCompletionAnnouncement
