I/O: Redesign how I/O handles averaging, fill values and average tracking

A part of the I/O restructuring will include how we handle output that needs memory of previous state. These would currently be AVERAGE, MIN and MAX.

A big part of this will be how we consider the average count flag. The average count flag is important in cases where a FillValue is present in some number of timesteps. We do not want to include FillValue values in the average computation and we want to make sure that when we ultimately average over X steps that X represents the total number of real valued contributions to the average.

Start by locating the I/O code that handles AVERAGE, MIN, MAX, fill values, and the average count flag. The issue provides no file or test names, so first map the current state-tracking flow. Done means fill values are excluded and the average count reflects real-valued contributions across the requested timesteps.

Written by the indexing model from the issue text.