[BUG] Inconsistent state after TSDataset.train_test_split

[d-a-bunin]

🐛 Bug Report

We have an inconsistent state after TSDataset.train_test_split if there were any transforms applied to dataset. Inconsistency is observed between columns and regressors.

Expected behavior

Regressors are subset of columns.

How To Reproduce

Code:

import pandas as pd
from etna.datasets import TSDataset
from etna.transforms import LagTransform


def main():
    df = pd.read_csv("examples/data/example_dataset.csv")
    ts = TSDataset(df=TSDataset.to_dataset(df), freq="D")

    transform = LagTransform(in_column="target", lags=[1, 2, 3], out_column="lags")
    ts.fit_transform([transform])
    print(f"TS features: {ts.columns.get_level_values('feature')}")
    print(f"TS regressors: {ts.regressors}")

    ts_train, ts_test = ts.train_test_split(test_size=24)
    print(f"TS-train features: {ts_train.columns.get_level_values('feature')}")
    print(f"TS-train regressors: {ts_train.regressors}")


if __name__ == "__main__":
    main()

Result:

TS features: Index(['lags_1', 'lags_2', 'lags_3', 'target', 'lags_1', 'lags_2', 'lags_3',
       'target', 'lags_1', 'lags_2', 'lags_3', 'target', 'lags_1', 'lags_2',
       'lags_3', 'target'],
      dtype='object', name='feature')
TS regressors: ['lags_1', 'lags_3', 'lags_2']
TS-train features: Index(['target', 'target', 'target', 'target'], dtype='object', name='feature')
TS-train regressors: ['lags_1', 'lags_3', 'lags_2']

Environment

No response

Additional context

No response

Checklist

• Bug appears at the latest library version