Don't show separate feature plots for training samples for pipelines with only pre-processing modules.

[damellis]

This is for pipelines that have only pre-processing modules but no feature extraction modules. In "feature mode" we should still only show one plot for each training sample, but it should be a plot of the pre-processed data. This follows our convention on the pipeline tab, where we show pre-processed data as a single plot and only feature data as separate plots.