JSON export incompletely specifies categorical splits of non-symmetric trees

Problem: JSON export incompletely specifies categorical splits (see below) catboost version: 1.2 Operating System: Mac OS X 13.3.1 CPU: Apple Silicon (M2) GPU: N/A

In the JSON export of a non-symmetric tree, the splits of categorical variables don't specify which categorical variable and prior are used for that split. Only the split_index is specified, and as far as I can tell, there is no way to use this to obtain the missing information.

Example:

import numpy as np
import pandas as pd
from catboost import CatBoostRegressor, Pool
import random

# Generate some random data
n_trees = 5
n_samples = 1000
n_float_features = 3
n_categorical_features = 2
n_cat_feature_values = 6

import string
def random_string():
    return ''.join(random.choices(string.ascii_uppercase, k=4))

cat_feature_values = [['feature_value_%i_' % j + random_string() for i in range(n_cat_feature_values)] for j in range(n_categorical_features)]

X = [list(np.random.rand(n_float_features)) + [random.choice(cat_feature_values[j]) for j in range(n_categorical_features)] for _ in range(n_samples)]
y = np.random.rand(len(X))

col_names = ['float_%02i' % i for i in range(n_float_features)] + ['categorical_%02i' % i for i in range(n_categorical_features)]
X_df = pd.DataFrame(X, columns=col_names)

cat_feature_names = [col for col in col_names if col.startswith('categorical_')]
                    
pool = Pool(
    X_df,
    label=y,
    cat_features=cat_feature_names,
    embedding_features=[]
)

model = CatBoostRegressor(
    iterations=n_trees,
    grow_policy='Depthwise',
)
model.fit(pool, eval_set=pool)

import json

fn = 'json-export-is-incomplete.json'
model.save_model(fn, format='json', pool=pool)
with open(fn) as f:
    json_data = json.load(f)

json_data['trees'][0]

This outputs

{'left': {'left': {'left': {'left': {'left': {'left': {'value': -0.0201874536772569,
       'weight': 3},
      'right': {'value': -0.035984302933017416, 'weight': 12},
      'split': {'border': 12.999999046325684,
       'ctr_target_border_idx': 0,
       'split_index': 127,
       'split_type': 'OnlineCtr'}},
     'right': {'left': {'value': 0, 'weight': 0},
      'right': {'value': -0.018084313720464706, 'weight': 1},
      'split': {'border': 2.9999990463256836,
       'ctr_target_border_idx': 0,
       'split_index': 80,
       'split_type': 'OnlineCtr'}},
     'split': {'border': 0.8814510107040405,
      'float_feature_index': 0,
      'split_index': 15,
      'split_type': 'FloatFeature'}},
...

Notice that splits of type FloatFeature have the float_feature_index. What is needed is the analog for splits of type OnlineCtr.