Improve the optimization of rho (of the FunDi model)

[trongnhanuit]

0. Set starting_rho = 0.5; converged = false;

1. Run a loop with up to 10 iterations
   1.1. Optimise rho from starting_rho and cbl from current branch length (of non-fundi model) to obtain fundi log-likelihood.
   1.2. If the resulting log-likelihood is not lower than non-fundi: converged = true; Stop.
   1.3. Reduce the starting_rho by half (i.e. 0.25 in the first iteration, 0.125 in the 2nd iteration, so on). Go back to step 1.1.

2. If not converged -> reset rho = 0 (non-Fundi model)

[bqminh]

all good, thanks Nhan!