Improve target handling

• Improved target handling in tasks, cli, datasets and predictionwriter for consistency, removing denominator_key from RegressionTask due to redundancy.
• Moved listify to array_utils, as I ended up using it in other parts of the code too.

Tasks remaining (all done):

• getting correct labels in predictionwriter : currently, in on_test_batch_end , they come out empty for some reason
• fix run_inference in RegressionTask
• actually remove denominator_key from RegressionTask
• Edit regression configs for tests of all types of regression tasks: single- and multi-output, with denominators, with norm_params, etc.
• (Unrelated) Add and extra check for number of regression task outputs >= number of targets.

Closes #35 (closed).