clamp periodic coordinate to prevent evaluation at rmax.

[Quntized]

The periodic wrapping formula in SplineEvaluator::eval() can produce

a coordinate exactly equal to rmax due to floating-point arithmetic. For periodic splines, rmax is outside the valid evaluation range [rmin, rmax) because rmax and rmin are the same physical point. Example: coord = rmin - 1e-16
floor((-1e-16 - 0.0) / length) = -1
coord -= (-1) * length = length - 1e-16 ≈ rmax

At rmax, eval_basis cannot find a valid cell , leading to incorrect B-spline evaluation.

Fix: after wrapping, if the result >= rmax, replace with rmin. This is mathematically equivalent for periodic domains (same point) but ensures eval_basis finds a valid cell.

[EmilyBourne]

Isn't t his case already handled by the tolerance in the eval_basis function? Do you have a MRE showing the problem you are fixing?

[tpadioleau]

@Quntized I agree with @EmilyBourne, could you provide a problematic snippet of code that your patch solves ?

[Quntized]

Thank you both for pointing this out. I checked get_icell_and_offset and find_cell_start — both already handle x >= rmax() by clamping to the last valid cell. I should have checked these functions before opening the PR. Lesson learned — I'll verify the full call chain before proposing fixes in the future. Closing this.