DefaultDatasetView::autoscale() is inefficient and broken

Sequential calls to `DefaultDatasetView#autoscale()` end up repeatedly computing the overall min and max. It could compute once and use cached results. Also it uses overall min and max rather than channel min and max which may be broken. We should revisit autoscaling.

Migrated-From: http://trac.imagej.net/ticket/1216

Start by locating DefaultDatasetView#autoscale() and trace how it computes the overall and channel min/max values. Confirm the desired behavior from the existing implementation and tests, then make repeated calls avoid redundant min/max computation and verify that scaling uses the appropriate channel values.

Written by the indexing model from the issue text.