Remove time_step global

[fweik]

~~and derived values~~. The time step should be parameter of the integration command.

[jngrad]

The time_step global is used in various places. Here is the list with proposed solutions:

• integrators
  • passing the time step by argument is trivial
• the Correlator class constructor
  • add a callback mechanism for observables: changing the time step during accumulation should throw an error
  • for example, we could store a shared pointer to the observables
• the thermalized_bond_set_params setter
  • set it to some undefined value: the thermalized_bond_init() callback iterates over all bonded interactions to update thermalized bonds with the current time step
• the DPD thermostat
  • set it to some undefined value: the dpd_init() callback updates the DPD struct with the current time step
• LB GPU and EK GPU
  • the time step needs to be copied to the GPU parameter structs
  • will be removed in 4.3
• LB CPU and IBM
  • it should be possible to pass the time step by argument through the callback chain, with enough patience
  • will be removed in 4.3
• H5MD

As a first step in 4.2, we could gather the integrator global variables (sim time, time step, force cap, etc.) into a global struct IntegratorState. The current integrator could be stored in a variant field of that struct, so as to avoid passing the time step by argument in too many places. This would also remove numerous switch statements.

[jngrad]

This was achieved by #4820. Closing.