fix: Avoid deep copying in ModelContainer to resolve pickling errors

[laksi1999]

Issue

The original implementation of the ModelContainer class relied on deep copying experiment objects stored in the model_object attribute. This approach led to "cannot pickle" errors when dealing with certain objects that are not serializable, such as those from the Driverless AI. This limitation hindered users from integrating their models seamlessly, particularly when using tools that are popular in the community.

Proposed Solution

• Introduces a new model_objects parameter, allowing users to pass pre-created model objects directly into the ModelContainer.
• The primary change involves,
  • Eliminating the deep copying of the model_object.
  • Instead of creating a deep copy, the code now checks for the availability of a model_objects list that contains pre-created instances.
  • If this list is populated, the method pops the first object from the list and assigns it to the model_object.
• This approach effectively resolves issues with non-picklable objects by ensuring that instances of ModelWrapper are created and passed directly, thereby avoiding serialization problems.

Note: Importantly, the existing logic for using a single model_object remains intact, allowing users to continue using the ModelContainer as they did previously without any required changes.