Reconsider allocation time stepping defaults for performance

[SouthEndMusic]

Given the findings here it might be worth reconsidering some of the default values for allocation configuration options that pertain to the timestep size:

• Whether adaptive time stepping should be the default
• Whether reltol_linearization can be increased

To answer this we must have our priorities for the allocation timestep clear:

• Do we ideally want to have a large timestep of say 1 day as a default because that best represents human behavior?
• Do we want smaller timesteps to reduce the physics errors made in the allocation solve?
• Do we want smaller timesteps for stability? The linked results seem to indicate that that is less of an issue now after multiple bug fixes.

[gijsber]

Stability was a major reason why adaptive time stepping was introduced. Better physics (lower linearization errors) were contributing to this goal. From a modeller's perspective a 1 day timestep would be a logical choice as it represents human behaviour best.
Maybe we should fix the default at 6 or 12h such that the linearization error is a bit sized down. Generally I am not very concerned on the linearization error since we keep track of actual physics in the physical layer.