Do not overwrite values from model configuration with defaults from user_configuration

There is a problem with user_configuration overriding when we use generic workers.

1. The model configuration has a value value1 for parameter param1
2. The user configuration specified in the worker run's configuration has a default value value2 for this parameter
3. The user doesn't specify any value for parameter param1
4. The user_configuration attribute is loaded with value value2 for parameter param1
5. As the user_configuration is always the last dict to override the configuration, at execution time, param1 has value value2 and not value1 as the ML engineer intended.