JSON dumps don't have split type

Function `XGBoosterDumpModel` has an option of dumping to JSON. This JSON output has a field `split_condition` which tells against what to compare the column in field `split`.

The JSON however doesn't mention how this `split_condition` is to be applied - for numeric features, it will be a threshold used as `split < split_condition`, while for categorial features it will use an 'in' condition `split in split_condition`.

The text and 'dot' dump formats both indicate what the split type is.

Start at the XGBoosterDumpModel JSON output path and compare how the text and dot dump formats represent split types. Verify the numeric threshold and categorical “in” cases described in the issue; done means the JSON identifies how split_condition is applied for both feature types.

Written by the indexing model from the issue text.