Add Flexible Switching Schedule Between Models

[Tetracarbonylnickel]

This PR introduces a modular and configurable framework for switching between two models during a simulation based on customizable conditions and transition schemes.

Key Features

• SwitchSchedule class:
  • Encapsulates logic for triggering and executing the model switch.
  • Uses a two-step strategy:
    • A condition function (CONDITION_FN) to determine when to switch.
    • A switching function (SWITCHING_FN) to compute the transition factor over time.
• Switching conditions (CONDITION_FN):
  • additive_impact: triggers switching when the last atom’s z-position drops below a given impact_height.
  • instant_switching: enables immediate switching (useful for testing or benchmarking).
• Switching functions (SWITCHING_FN):
  • linear: linearly interpolates the switching factor from 0 to 1.
  • sigmoid: smoothly interpolates the switching factor using a logistic sigmoid ramp (parameterized by k).

Motivation

Switching between ML models (e.g., triplet → singlet) often requires:

• Careful handling of when to switch
• Smooth transitions to avoid unphysical forces
  This design supports plug-and-play extension for new switching conditions or ramping strategies.

TODO

• Handling auxiliary energy and force functions of the two models to ensure correct energy forces and their uncertainties.