Remove dt from Timestep

[yuxuanzhuang]

As per discussion in #2723 (comment).

There seems to be limited use of Timestep.dt but raising issues e.g. have to define self._reader as weakref for some readers. Besides, it does make to much sense to have dt defined for each Timestep (maybe not the case for a multiple timestep simulation? Also not sure we can/cannot handle it now.); we can easily have trajectory handling time difference itself. What's your opinion?

[orbeckst]

Conceptually it does not make sense to associate a time difference with a point in time so I am very much in favor of removing it, especially when it simplifies other code and decouples Timestep from a Reader.

(There are trajectory formats (such as TRR) that can have variable and arbitrary dt so for these it might not even make sense to have a trajectory.dt but that’s a different problem.)

[orbeckst]

Please note that if we remove Timestep.dt then we need to add a deprecation to the 1.0.1 backport PR #2798 :

1. Create one PR (against develop, as usual) to remove Timestep.dt
2. Create a second PR (against PR Backport 1.0.1 #2798 or against master) to add the deprecation.