Should coarse data be warmed up with coarse or reference stepper?

[Ceyron]

Only relevant for scenarios that are warmed up, i.e., that have num_warmup_steps > 0. These are typically chaotic problems that are warmed up to have the emulators be trained on the attractor.

Currently, the scenario methods get_train_data_coarse() and get_test_data_coarse() use a warm-up with the coarse stepper.