fix(behaviors): Fixing erroneous combo triggering, hold-taps sticking

* This is a very simple fix to a rather complicated issue. Essentially,
hold-taps will "release" (raise) their captured keys before actually
telling the event manager they have captured a key. This means the event
manager ends up assigning the `last_listener_index` to the hold-tap
subscription rather than the combo. So when the combo calls
`ZMK_EVENT_RELEASE` it raises after the hold-tap instead of after the
combo as the combo code expects.
* The corresponding test (which fails without this change) has also been added.
* An event can be captured and released in the same event handler, before
the last_listener_index would have been updated. This causes some handlers
to be triggered multiple times.
* The solution is to update the last_listener_index before calling the next
event handler, so capturing and releasing within an event handler is harmless.
* Also see discussion at https://github.com/zmkfirmware/zmk/pull/1401
* If our handler dedides our undedided hold-tap,
  return early before continuing.
* Fix incorrect pointer logic, resulting in combo
  candidate filtering leaving incorrect timeout details.

Co-authored-by: Andrew Rae <ajrae.nv@gmail.com>
Co-authored-by: okke <okke@formsma.nl>

app/src/combo.c

@@ -211,7 +211,7 @@ static int filter_timed_out_candidates(int64_t timestamp) {
         if (candidate->timeout_at > timestamp) {
             // reorder candidates so they're contiguous
             candidates[num_candidates].combo = candidate->combo;
-            candidates[num_candidates].timeout_at = candidates->timeout_at;
+            candidates[num_candidates].timeout_at = candidate->timeout_at;
             num_candidates++;
         } else {
             candidate->combo = NULL;