… failures

* feat: add UI element detection to snapshots

Add element detection for identifying interactive UI elements in terminal
screens. The system uses visual style segmentation and pattern
classification to identify buttons, inputs, and toggles.

Key components (crates/pilotty-core/src/elements/):
- grid: maps vt100 screen to typed cells with style info
- segment: clusters adjacent cells by visual style
- classify: pattern-matches clusters to element types
- Element: output type with kind, position, text, confidence

Detection patterns:
- Buttons: [OK], <Cancel>, inverse-video text, underlined text
- Inputs: underscore runs (____), cursor-containing regions
- Toggles: [x], [ ], (o), ( ) checkbox/radio patterns

Classification priority ensures pattern recognition takes precedence
over cursor position (cursor on button = focused button, not input).

Elements include row/col for direct use with click command.
Confidence scores indicate detection certainty (1.0 = high, 0.6 = low).

Known limitation: ncurses apps that style brackets differently from
interior text (e.g., dialog) may not detect buttons correctly.

* docs: readme