Add support for nested variogram models.

[unageek]

Phase 1

Integrate anisotropy into RBF definition.

• [x] Add an affine transformation which transforms the RBF to rbf_base
• [x] Change the parameter type of rbf_base::evaluate from double to const geometry::vectors3d& (still translation invariant)
• [x] Do affine transformation of points in the following classes
  • fmm_evaluator
  • fmm_operator
• [x] Remove transformation from interpolant

Phase 2

Move the nugget parameter from RBF to model.

• [x] Add nugget parameter to model and remove from RBFs
• [x] Modify variogram_fitting to take model instead of covariance_function_base

Phase 3

~Remove the range parameter from covariance functions (which can be specified by a transformation)~

• [ ] ~Remove range parameter from covariance_function_base~
• [ ] ~Modify variogram_fitting~

See my comment in #62.

Phase 4

Extend model to support multiple RBFs.

• [ ] Extend model class to take multiple RBFs But keep the single-RBF ctor for convenience.
• [ ] Add cpd_order() function to model which returns the maximum order of CPD of the RBFs
• [ ] Modify variogram_fitting

RandomFields has RMplus for this purpose.