xgboost and categorical predictors

[topepo]

There is an experimental argument for categorical variables.

We need a process to potentially adopt this and how we should handle

• backward compatability
• changes to parsnip:::as_xgb_data() since some new xgboost options are required for this.

[EmilHvitfeldt]

Related issues in xgboost

• [R] Implement feature info for DMatrix. dmlc/xgboost#8048
• Categorical data support (part 2) dmlc/xgboost#7899

[exsell-jc]

I thought I created an issue about this, but I might be wrong -- might've been in a wrong repository.
It would be really nice if this gets implemented.

[simonpcouch]

Related to #1087. Once the xgboost team starts moving toward an R release of xgboost 2 we can do a push on supporting those new features and interfaces!