Generalize input module to mouse and joystick/gamepad events

Extend the input service beyond the keyboard so mouse and joystick/gamepad
drivers can broadcast too, with per-device publish and subscribe methods.

- protocol: KeyEvent joins MouseEvent (motion/buttons/scroll) and
  JoystickEvent (axes/buttons), all carried in a common InputEvent envelope
  tagged with a DeviceKind. A subscribe request carries a device_mask, so a
  subscriber names the classes it wants and the service routes each event only
  to interested subscribers (a mouse-only listener never wakes for keystrokes).
- runtime: per-device publish methods (publishKeyboardEvent/publishMouseEvent/
  publishJoystickEvent) and subscribe helpers (subscribeKeyboard/Mouse/Joystick,
  each typed, plus subscribe(mask)/subscribeAll returning the tagged envelope).
- service: subscriber table gains a device_mask; broadcast routes by the
  event's device class.
- mouse driver now publishes (synthetic) mouse events like the keyboard driver;
  input-source cycles all three classes; input-test subscribes to all and only
  emits its "ok" marker once it has received one of each class — so the passing
  test proves per-device routing, not just delivery. Real HID decoding stays a
  follow-up.

No kernel changes: ipc_send is generic and the 36-byte InputEvent fits its
64-byte payload. Full QEMU suite 48/48; serial log confirms keyboard, mouse,
and joystick all reach one subscription.

docs/README.md

@@ -52,9 +52,10 @@ rather than restate it. Roughly in the order things happen at runtime:
     microkernel's `ps`/`kill`/SIGCHLD: enumerate as a table snapshot, the
     supervision link as the kill authority, and child-exit notifications over the
     same endpoints IRQs arrive on.
-16. **[input.md](input.md) — the input module.** Broadcasting keyboard events: why a
-    synchronous rendezvous can't fan out to many listeners, the asynchronous `ipc_send`
-    primitive built to fix it, and the subscribe/publish service layered on top.
+16. **[input.md](input.md) — the input module.** Broadcasting input events (keyboard,
+    mouse, joystick): why a synchronous rendezvous can't fan out to many listeners, the
+    asynchronous `ipc_send` primitive built to fix it, and the per-device subscribe/publish
+    service layered on top.
 17. **[halting.md](halting.md) — halting.** Why a kernel can't just "exit", and
     how `while (true) hlt` parks the CPU safely once there's nothing left to do.