[MNT] Refactor `run_flow_on_task`-function

[SimonBlanke]

The run_flow_on_task is ~160 lines of code with high complexity: https://github.com/openml/openml-python/blob/main/openml/runs/functions.py#L178

I would suggest to refactor it to reduce its complexity. We can keep the this main-function as the orchestrator and separate the responsibilities into:

• input validation
• parallel processing
• aggregation of predictions and evaluations
• creating the run from the res