Fix/alp time loop termination

[philipphaslhofer]

Summary

The ALP pulse-time loop had two bugs:

1. Broken termination condition. while (std::fabs(time - pulseTime) > 1e-6) only exits
   when time lands within 1e-6 of pulseTime. For any pulseTime / coverageTimeStep that
   is not an exact integer, time overshoots and the fabs keeps growing — the loop never
   exits. Only the "neatly divisible" case worked.
2. Last-step overshoot. The final iteration runs fluxes and
   coverage updates with the full coverageTimeStep, so the physics integrates over more time
   than actually remains in the pulse.

Changes

• Commit 1 — replace the fabs guard with a monotonic time < pulseTime - coverageTimeStep * 1e-4 check so the loop terminates for arbitrary pulseTime / coverageTimeStep ratios and is invariant of the time scale.
• Commit 2 — clamp dt = min(coverageTimeStep, pulseTime - time) and push it to the
  surface model via a new setTimeStep hook on SurfaceModel, so the final iteration
  integrates over the actual remaining time. SingleParticleALDSurfaceModel overrides the
  hook to update its (now non-const) dt_; base-class default is a no-op so other models are
  unaffected.