Add cubic and RBF interpolation for adjoint inversion

[cpjordan]

Adds adjoint-safe independent-point interpolation options for the headland inversion example.

Changes

• Added IndependentPointControlMapping and IndependentPointInterpolator to thetis.inversion_tools.
• Supports independent-point mappings with:
  • linear
  • cubic via SciPy CloughTocher2DInterpolator
  • rbf via SciPy RBFInterpolator
• Uses Firedrake external operators for non-UFL interpolation so pyadjoint can propagate gradients.
• For cubic interpolation, uses a value-dependent finite-difference vector-Jacobian product because precomputed identity basis masks do not exactly reproduce SciPy Clough-Tocher interpolation on the headland mesh.
• Updated examples/headland_inversion/inverse_problem.py to expose:
  • --ip-interpolation {linear,cubic,rbf}
  • --ip-rbf-kernel
  • --ip-rbf-smoothing
• Updated the headland Makefile with:
  • IP_INTERPOLATION
  • IP_RBF_KERNEL
  • IP_RBF_SMOOTHING
• Updated plotting script to use matching output directories for cubic/RBF runs.
• Updated README documentation for the new independent-point interpolation options.
• Added adjoint regression tests for linear, cubic, and RBF independent-point mappings.