[BUG] Something went wrong with multi-index if backtest

[julia-shenshina]

🐛 Bug Report

I ran code from How To Reproduce and got strange results: fold_info column in backtest's forecasts was not added to ts[:, segment] block but concatenated separately from other segment's columns.

Expected behavior

I expected smth like

|segment |     segment_0      |     segment_1      |      segment_2     |
-------------------------------------------------------------------------
|feature | target | fold_info | target | fold_info | target | fold_info |

How To Reproduce

Code

from etna.datasets import TSDataset, generate_ar_df
from etna.ensembles import VotingEnsemble
from etna.pipeline import Pipeline
from etna.models import NaiveModel
from etna.metrics import MAE

df = TSDataset.to_dataset(generate_ar_df(periods=100, start_time="2020-01-01", n_segments=3))
ts = TSDataset(df, "D")

HORIZON = 5
ensemble = VotingEnsemble(
    pipelines=[
        Pipeline(model=NaiveModel(lag=7), transforms=[], horizon=HORIZON),
        Pipeline(model=NaiveModel(lag=10), transforms=[], horizon=HORIZON),
    ]
)

_, forecasts, _ = ensemble.backtest(ts, metrics=[MAE()])
forecasts

Screenshots

Environment

etna==1.5.0

Additional context

No response

Checklist

• Bug appears at the latest library version