composition API, which is used to select the "most applicable" model
(Trade Condition, Transformation, Pay Sheet Model, etc) with the same
reference was only discarding deleted and validated models, but it
was selecting draft models.

This leads to this kind of problem:
 - A validated model exist, with an effective date
 - A new model is made by cloning the validated one, edited to set
  effective date from now and validated
 - The first validated model is also cloned by mistake, but not
  validated
In such case, the model cloned by mistake might be used, which from
user point of view is wrong, because user validated a model, so if
the system use a draft instead, that's a very wrong behaviour.

This change require models to be validated for the composition API to
work, many tests were updated to validate the model (either in the test
or in the data business template used by the test).

This change might also produce different results in production, because
it was possible that some draft models were selected before, now they
will no longer be selected. To check for potential side effects, we
recommend looking for all draft models with a reference, examine them
one by one and validate the model if it's a case where user forgot to
validate.