Define REANA workflow ML scripts

[Sinclert]

This PR performs a big refactor on how the shell script commands are run on the REANA workflow. The main goal of the refactor is to move all the shell script commands from the steps.yml file into proper .sh files, so that they can be run locally for debugging purposes.

This PR covers the "ml" use-case steps.

---

During the refactor, additional modifications have been performed:

• Update the Docker image to copy the brand new scripts.
• Define a new ML Makefile rule: clean.
• Define a save_results function within the evaluation.py module, in order to simplify saving operations at the end of the module.
• Replace h5py context managers by common Python file opening (here and here). If a context manager is used, data within the .h5 file is locked the moment the execution flow goes out the context... 🤦
• Replace hand-coded Fisher thetas by their arguments version.
• Replace the evaluate_log_likelihood function by evaluate_score when evaluating a sally or a sallino method. (Following what is specified in the official tutorial).
• Delete unnecessary shuffle option from the reana/inputs/input.yml file.
• Delete reana/inputs/input.yml extra parameters causing problems within "alices" and "alice" methods (here and here).
• Delete a "half-way-done" evaluation section within reana/inputs/input.yml.

---

New additions:

• Evaluation results are now differentiated by their method:
  • Discovering their differentiated names, here.
  • Told REANA that there will be several of them, here.
  • Iterate over them in the plot step, here.

[Sinclert]

If all changes seem OK, this PR is ready to merge.