Fix BSpline eval_unstructured for multi-dimensional interpolation

[ChrisRackauckas-Claude]

Summary

• Fixed a bug where eval_unstructured was failing for multi-dimensional BSpline/NURBS interpolation
• Added regression tests for BSpline and NURBS eval_unstructured
• Verified interface compatibility with BigFloat, Float32, and standard arrays

Problem

The eval_unstructured function was throwing a BoundsError when used with multi-dimensional BSpline interpolation (e.g., 2D or higher). For example:

t1 = [-3.14, 1.0, 3.0, 7.6, 12.8]
t2 = [-2.71, 1.41, 12.76, 50.2, 120.0]
t1_eval = t1[1:(end - 1)] + diff(t1) / 2
t2_eval = t2[1:(end - 1)] + diff(t2) / 2

u_bspline = fill(2.0, 6, 7)
itp_bspline = NDInterpolation(
    u_bspline,
    (BSplineInterpolationDimension(t1, 2; t_eval = t1_eval, max_derivative_order_eval = 1),
        BSplineInterpolationDimension(t2, 3; t_eval = t2_eval, max_derivative_order_eval = 1))
)

result = eval_unstructured(itp_bspline)  # This was throwing BoundsError

Root Cause

In eval_kernel, when eval_grid=false, @index(Global, NTuple) returns a 1-tuple (e.g., (1,)). This was passed directly to _interpolate! as multi_point_index. However, the BSpline get_basis_function_values method for tuples expects an N_in-tuple and tries to access multi_point_index[dim_in] for all dimensions, causing a bounds error when dim_in > 1.

Fix

Convert the 1-tuple to a scalar using only(k) before passing to _interpolate! when eval_grid=false. This ensures the correct method dispatch:

• Scalar Number → unstructured evaluation method
• N_in-tuple → grid evaluation method

Test plan

• All existing tests pass
• New tests added for:
  • BSpline 2D eval_unstructured
  • BSpline eval_unstructured with BigFloat
  • NURBS eval_unstructured

Interface Compatibility Check

This PR also checked interface compatibility:

• ✅ BigFloat works correctly with all interpolation types
• ✅ Float32 works correctly with type preservation
• ✅ ArrayInterface compatible (uses similar(), zero(), one() correctly)
• ⚠️ JLArrays: fails due to missing synchronize for JLBackend in JLArrays (this is a JLArrays issue, not DataInterpolationsND)

cc @ChrisRackauckas

🤖 Generated with Claude Code