Some default priors are not very good

[j-faria]

Now that (almost all) the default priors are in one single place (default_priors.cpp), it would be good to check if they are ok.
In particular, some ModifiedLogUniform priors use somewhat arbitrary values for the knee parameter, which can lead to unexpected behaviour.

[j-faria]

I'm thinking that replacing the ModifiedLogUniform priors with HalfGaussians could solve this.

This could change the default priors for

• the jitter (Jprior)
• the $\eta_1$ parameter in the GPmodel
• the corresponding parameters for the activity indicators

[j-faria]

Alternatively (for the jitter), set the knee of the ModifiedLogUniform to be the average uncertainty in the timeseries