* Move to local/stack allocated event API that doesn't require
dynamic allocation/freeing.
* Disable heap, we no longer use alloc/free unless using LVGL.
* Tons of refactors all over to account for the new event approach.
Added BEHAVIOR_DT_DEFINE() and BEHAVIOR_DT_INST_DEFINE(), which work
exactly like the DEVICE_*_DEFINE() macros, except they also register the
device as a behavior by adding a pointer to it to a memory section.
Added zmk_behavior_get_binding(), which works like device_get_binding()
except that it only searches the devices that have been registered as
behaviors. This ensures that behaviors cannot have name collisions with
other devices defined by the SoC, which will be important when we remove
the label property from behaviors so they are given their node names.
As an added benefit, this is faster since it searches a smaller list.
Some basic benchmark code I wrote indicates it takes 30-70% as long,
depending on where the behavior is in the list and whether the name
string is an exact pointer match.
From now on, behaviors should use BEHAVIOR_*_DEFINe() instead of
DEVICE_*_DEFINE(), and any code that looks up a behavior by name should
use zmk_behavior_get_binding() instead of device_get_binding().
