More flexible bayesian Royston-Parmar models.

[albertocarm]

We need the Royston-Parmar Bayesian model to be more flexible. At present, the fit.models function only allows to adjust the k parameter but does not allow to model the dynamic effect of covariates. For example, with the flexsurvspline function, a more flexible model would be specified with the formula flexsurvspline(Surv(time,event)~arm+gamma1(arm)+gamma2(arm)+gamma3(arm), data=data) However, at present, fit.models does not accept this formula. It would also be useful to incorporate a new function to plot: (1) the difference in survival rates over time with the correponding credible interval, (2) the hazard ratio over time with its credible interval, (3) a function to test bayesian hypotheses over time (e.g, whata is the posterior probabilty that HR is >1.15 at time=12 months? or is the difference in OS rates >0 at 24 months?).