fix: preserve final timestep when resampling trajectories

[kejan2514]

Summary

Fixes trajectory resampling so that the final timestep is always preserved.

Previously, np.arange() could exclude the trajectory endpoint when the time range was not evenly divisible by target_dt.

Changes

• Preserve the exact final trajectory timestamp during resampling
• Avoid duplicate endpoints when target_dt divides the interval exactly
• Add regression tests for both divisible and non-divisible intervals

Testing

Added tests covering:

• Exact interval division
• Non-divisible intervals with a shorter final step

[faisalnugroho]

Reviewed the patch — I reproduced the bug on main (np.arange(0, 1, 0.1)[-1] == 0.9) and verified this fix against several edge cases:

• ✅ Span divisible by dt → grid ends exactly at t_end (11 points for 0–1s @ dt=0.1)
• ✅ Span not divisible by dt → exact landing with a shorter final interval (e.g. [0, 0.3, 0.6, 0.9, 1.0])
• ✅ Single-frame trajectory (t_start == t_end) → grid collapses to one point instead of being empty
• ✅ Float drift case (dt = 0.1000001) → isclose catches the near-final point and snaps it to t_end

One minor nit: in the else branch, new_times[-1] = t_end only triggers when the last point is already within isclose tolerance of t_end, so it's effectively a snap-to-endpoint — correct, but worth a one-line comment since future readers may assume it's a no-op.

The included regression tests cover the two main behaviors nicely. From my side this looks safe to merge — leaving the maintainer to confirm the design choice (strict-uniform vs exact-landing), which the linked issue #2 asks about.