ASE backend: avoid duplicate sampling updates within a single MD timestep

[Songze1019]

Problem

In the ASE backend, Sampler.get_forces() may be called multiple times within a single MD timestep.
Previously, each call triggered:

• snapshot refresh
• method_update(...)
• callback logging
  For metadynamics, this caused state.ncalls to advance faster than actual MD steps, leading to over-frequent hill deposition and mismatched hill step indexing.

Root Cause

pysages/backends/ase.py used force-call frequency as the effective update clock.
In ASE/Langevin workflows, force calls can occur multiple times per timestep.

Fix

In pysages/backends/ase.py, gate method updates by ASE MD timestep:

• track self._last_update_timestep
• run snapshot/update/callback only when timestep changes
• still return forces + bias for every force request
  This preserves force application while preventing duplicate state advancement in the same timestep.