Pierrick Martin / ctools

…e error is larger than the value (#2762).

Setting the parameter error to the parameter value is a problem to cases where the parameter value is zero. Also, we do not know why the error is larger, this can be completely normal for sources with negligible flux.

All likelihood ctools now derive from the ctlikelihood base class. This avoids code duplication and provides all likelihood tools with a uniform interface.

…gning the old best optimizer result).

… csspec and complemented the copy constructor of cterror (the optimiser was not copied and intialised)