Don't treat composite HID keyboards as game controllers (#1839)

* Don't treat composite HID keyboards as game controllers

isGameController() calls getMotionRange() without a source class, so any
device that claims SOURCE_JOYSTICK in its source mask counts as a controller
as soon as it reports AXIS_X/AXIS_Y through *any* source.

Bluetooth keyboards with a built-in touchpad hit exactly that: the touchpad
reports those axes under SOURCE_MOUSE, while the composite HID descriptor also
advertises SOURCE_JOYSTICK. The keyboard then gets bound as a virtual XInput
pad, which takes over library navigation and leaves keyboard+mouse unusable in
game.

Query the motion ranges under SOURCE_JOYSTICK explicitly, in both copies of
the check, matching what PhysicalControllerHandler.kt already does. Real
controllers are unaffected: they either report joystick axes under
SOURCE_JOYSTICK or match on the (isGamepad && hasGamepadKeys) branch, which
this does not touch.

* Also accept controller axes reported under SOURCE_GAMEPAD

Review feedback: the joystick-only query is narrower than the helper in
PhysicalControllerHandler.kt, which tries SOURCE_JOYSTICK, then SOURCE_GAMEPAD,
then the source-agnostic overload. Query SOURCE_GAMEPAD as well, so a driver
that attaches the sticks to the gamepad source instead of the joystick source
keeps being classified.

The source-agnostic third arm is deliberately left out: that is the one that
matches a touchpad's SOURCE_MOUSE axes, i.e. the misclassification this change
exists to fix. It is safe in PhysicalControllerHandler because hasMotionRange
is only reached from deviceHasTriggerAxis for a device that already has a
controller profile bound, never to decide whether a device is a controller.

* Keep a single copy of the axis-source check

Review feedback: the new helper was pasted into both classes. Define it once in
ExternalController (package-private) and have ControllerManager call it, so this
change adds no new duplication.

Merging the two isGameController() copies outright is left out on purpose: they
have drifted, ExternalController skips virtual devices and ControllerManager
does not, so unifying them means either starting to skip virtual devices when
assigning controller slots or stopping to skip them in the xserver path. That is
a behaviour change beyond this bug fix and wants its own PR and a device to test
on.