refactor feature importance code

[hbaniecki]

1. rename permutation to _permutation_, times to _times_ and reference to _reference_ to omit potential problems with feature names

   survex/R/surv_feature_importance.R

   Line 160 in 64e96d6

   raw$permutation <- rep(1:B, each = length(times))

   
   

   survex/R/surv_feature_importance.R

   Line 156 in 64e96d6

   cbind(data.frame(times = times), "_full_model_" = loss_full, loss_variables, "_baseline_" = loss_baseline)

2. use ret[['_reference_']] instead of ret[, ncol(ret)] for readability and code robustness

   survex/R/surv_feature_importance.R

   Line 175 in 64e96d6

   res[, 2:(ncol(res) - 3)] <- res[, 2:(ncol(res) - 3)] / res[, (ncol(res))]

3. sort the result by _permutation_ after merge() like it is done in type="raw"

   survex/R/surv_feature_importance.R

   Line 170 in 64e96d6

   res <- merge(res, res_full, by = "times")