Wildcard callbacks during app initialization behave improperly

[Marc-Andre-Rivet]

Using dash==1.12.0

Looking at the dash-renderer test cbmo003(https://github.com/plotly/dash/blob/dev/tests/integration/callbacks/test_missing_outputs.py#L187), the initialization behavior seems to be incorrect.

The callbacks are:
cb1. items.n_clicks -> (content1.children, content2.children) , children of content1 are of shape {i: number} , children of content2 are of shape {j: number}
cb2. values.n_clicks -> ({i:ALL}.children, {j:ALL}.children)
cb3. ({i:ALL}.children, {j:ALL}.children) -> output.children

At initialization, cb1 and cb3 are triggered

• cb1 returns 2 children from content1 and none for content2, the children of these wildcard components are all null
• cb1 result processing triggers cb2 because there’s now != 0 wildcard outputs
• cb3 is run with children values ([None, None], []) and returns 0because of value or 0
• cb2 is run with None -> ([0, 0], [])

Expected behavior
We trigger cb3 again after assigning the children values after cb2 , similar to the followForward cases. The problem becomes more apparent when changing cb2 to return, for example, 5 instead of 0 on the None case — the test now displays the screenshot below, showing that it never really used the children content to calculate itself during initialization.