GTR substitution model rates should default to a Dirichlet prior in beauti

[jordandouglas]

Currently, the beast.base.spec.evolution.substitutionmodel.GTR class is made available through BEAUti. This class accepts six individual rates as an input (rateAC, rateAG, ...). These rates are IID but one of them is fixed to 1 so they are identifiable.

I suggest this class is removed from the BEAUti templates entirely and replaced with beast.base.spec.evolution.substitutionmodel.GeneralSubstitutionModel, which takes on a vector of rates. These rates should have a Dirichlet distribution by default, with shapes that reflect our prior expectation that transitions are more common than transversions, for example a default of Dirichlet(rAC=2, rAG=4, rAT=2, rCG=2, rCT=4, rGT=2). This parameterisation is more natural and we should take this chance to move away form the previous one.

[rbouckaert]

This is how it is implemented in the SSM package.

To keep the same interface in BEAUti (allowing individual rates to be clamped to a user defined value) would require use of a CompoundRealScalarParam. Not quite sure what follow up issues that would bring.