Extensisq

Stiffness detection for Runge Kutta methods is added, and the methods are given an simpler name.

Other changes to the Runge Kutta base class:
- The stepsize near the end of the integration is different:
- look ahead to prevent too small step sizes
- linear extrapolation if the last step is too small despite it.
- the min_step accounts for the distance between C-values
- the scale (weight) is smoothed differently
- a different tolerance validation is used.

Most of these changes originate from the RKSuite code.

The Fortran code `DDEABM` is translated to Python and impemented as an `OdeSolver` class for `solve_ivp`.

Other updates:
* The error estimator order (instead of method order) is used for the initial step size estimation of the Runge Kutta methods.
* minor changes

Updates:
* Add more elaborate intial step estimator (by H.A. Watts)
* Add linear and cubic dense output classes
* Custom RungeKutta class that differs from Scipy's class in:
* intial step size method
* handling of end of the integration range (t_span[1])
* the dense output classes
* includes failed step counter

Changelog:
* High order Prince methods added
* Jupyter notebook for Prince methods
* Created a base class for conventional RK methods
* Created a base class for dense output with Horner's rule
* Some minor simplifications and modifications

package license for conda forge

first release of extensisq