Cobyqa

This release improves the handling of the NaNs in the objective and constraint function evaluations.

1. The objective function values and maximum constraint violations in the history and the filter are now those without extreme barriers.
2. If the returned objective function value or the maximum constraint violation is NaN, the optimization procedure is now considered unsuccessful. It is important to note that this is only possible if the method does not find any point with well-defined objective and constraint function values. If any point with well-defined objective and constraint function values is encountered during the optimization procedure, then the returned values in `fun` and `maxcv` are not NaN.

1. The computations of the quadratic models have been improved. Instead of using an LBL factorization to solve the KKT conditions, we now employ an eigendecomposition-based method, improving the stability of COBYQA.
2. Passing unknown constants to the `minimize` function now raises a warning.

This release improves the returned value of the `minimize` function and corrects a bug where nonlinear constraints could not be passed as a `dict`.

This is a bugfix release.
1. The documentation has been improved.
2. Typos in the examples have been fixed.
3. Constants can now be modified by the user.

This is the initial release of COBYQA for Python.