Resolve "Training initial population necessitates .mll config and pretrained file"

The initial population provider training that uses training data or already known pairs of search and optimisation values, e.g. training.json, for the initial population was broken. This commit fixes it. Using the training provider is possible without specifying a model configuration and a learned surrogate model. There are two ways to configure the provider:

1. Using only the .ol file

Please use the following configuration snippet and adapt the filename according to your needs:

  ...
  initialisation := training {
                      'training-file' := "training.json";
                    };
  ...

2. Using the .ol file and the blackboard configuration

Please use the following configuration snippet:

  ...
  initialisation := training {};
  ...

In this case, the filename is passed by the command-line parameter -Bsurrogate:training-data=training.json.

Closes #113 (closed)

Changelog: bug Changelog: changed