They should use gpa.destroy() instead of foo.destroy() because (most) of
them have fields that may not be initialized by the first error, so
the foo.destroy() could crash.
We need to defer config application to the first manage_start event
using a should_manage flag so that all *_support events have arrived
before we try applying the configs
This commit also has two other fixes
- fixes a potential use-after-free by telling InputDevice when a
LibinputDevice is .removed.
- fix logFn (removed "if (scope != .default) return;")
I used kwm to help figure out the manage pattern for the input config.
Link to kwm: https://github.com/kewuaa/kwm
Right now, the support is still incomplete (no way to set config) but
we get the devices and set them up and handle current/support events
for the river_libinput_device_v1 devices.
