feat: thread device identity for per-device switch conditions

[malpern]

Summary

• Adds device_index: u8 field to KeyEvent, threaded from Linux KbdIn through to keyberon switch evaluation
• On Linux, each registered input device gets a monotonically increasing index assigned at registration time
• New (device N) switch condition in the parser that matches events from a specific physical device
• Stores current_device: Option<u8> on Layout (not in keyberon Event row) to preserve the row 0/1 real/virtual key invariant
• macOS/Windows compile cleanly with device_index: 0 default (no multi-device support yet)

Data Flow

Linux KbdIn::read()             -- Token → device_indices[token] → u8
  → Vec<(InputEvent, u8)>
  → event_loop: KeyEvent.with_device(idx)
  → tx.try_send(KeyEvent{code, value, device_index})
  → handle_input_event:
      layout.current_device = Some(event.device_index)
      layout.event(Event::Press(0, evc))   ← row still 0
  → Layout::do_action → Switch(sw)
      sw.actions(..., self.current_device)
  → evaluate_boolean:
      Device(idx) => current_device == Some(idx as u8)

Usage Example

(switch
  ((device 0) X break)
  ((device 1) Y break))

Not in scope (follow-up PRs)

• Name-based matching (device "Kinesis") / (defdevice ...) config
• Per-device layer stacks
• Windows/macOS multi-device support
• Device hot-plug stable indexing

Test plan

• cargo build succeeds
• All existing keyberon tests pass (79 tests)
• All existing parser tests pass (100 tests)
• All existing kanata lib tests pass (28 tests)
• New unit tests: device_opcode_encoding_roundtrip, device_evaluate_boolean_matches, device_evaluate_boolean_with_and
• Manual test on Linux with 2 keyboards using (device N) switch conditions

🤖 Generated with Claude Code

[malpern]

Closing — opened prematurely against upstream. Will validate locally in KeyPath integration before upstreaming.