TODO for version 1.0
- Transfer (some?) unit tests to examples to make them more easily accessible
  for the user - they will be tested during R CMD check anyway
- Let mkin call mkin_wide_to_long automatically, if observed data are suitably defined
- Think about what a user would expect from version 1.0
- Check the chi2 error calculation in mkinerrmin.R. KinGUII does this without iteration

Nice to have:
- Improve choice of starting parameters, in order to avoid failure of eigenvalue based solutions
  due to singular matrix
- Consider reverting to deSolve in case eigenvalue based solution fails due to singular matrix
- Calculate confidence intervals for DT50 and DT90 values when only one parameter is involved
- Calculate transformation only DT50 values (exclude pathways to sink) as additional information
- Calculate DT50 values from smaller rate constant/eigenvalue of DFOP, HS and SFORB models
- Document assumptions and implications of fitting method used in FME in the package vignette