Previously the K matrix model was intertwined with the BoxQuantizationclass,
e.g. fit parameters were accessed through BoxQuantization. Instead, the
BoxQuantization class should not care about how a K Matrix model is
implemented as long as it provides the two functions

	calculate(..)
	isZero(..)

now defined pure virtual in the new KtildeMatrixBase base class. A
K-matrix model is implemented by subclassing KtildeMatrixBase and
providing these functions. Model parameters are handled exclusively by
the K matrix model provided by the user.

A BoxQuantization object is now constructed by passing in an object
derived from KtildeMatrixBase and a boolean indicating whether this is a
model for Ktilde or KtildeInverse.

The example driver routines have been updated to reflect the changes in
interface.

This change in normalization induces a change in the normalization of
the Ktilde matrix to make it more natural. The difference is most easily
seen by comparing

	eq. (2.9) of https://arxiv.org/abs/1802.03100v1	(new convention)
	eq. (46)  of https://arxiv.org/abs/1707.05817v1 (old convention)

In the old normalization, the K matrix depended on m_ref L, i.e. a
finite-volume quantity. Using the new normalization, the K matrix does
not pick up a dependence on any finite-volume quantity, and the
Breit-Wigner parametrization takes its canonical form.