Configure sparse grid usage of fidelity indices

[eckelsjd]

Expected Behavior

• Should be possible to adjust how the SparseGrid converts fidelity indices to training data.

Actual Behavior

• Currently must use the assumed form of num_pts = k * beta + 1 in the beta_to_knots function. The scale "k" is configurable as self.knots_per_level.

Ideas

• Could allow a completely custom beta_to_knots function -- in which case you should just subclass SparseGrid and override the beta_to_knots function.
• Could easily add a configuration to SparseGrid for the offset, so that num_pts = k * beta + offset, where k and offset are both configurable. Anything more complicated should just override beta_to_knots. This would be useful for example, for setting the initial polynomial degree of Lagrange to something higher than 0.