recursive features not being created as expected

[jmoralez]

Describe the bug
When using lags the features in the first forecast step don't seem to consider the last observed value of the target, which should be the lag1 when making the first prediction.

To Reproduce

# %pip install utilsforecast
import polars as pl
from functime.forecasting import linear_model
from utilsforecast.data import generate_series
from utilsforecast.plotting import plot_series

serie = generate_series(1, max_length=50, engine='polars')
h = 14
train = serie[:-h]
valid = serie[-h:]
model = linear_model(freq="1d", lags=7, strategy="recursive")
model.fit(y=train)
y_pred = (
    model.predict(fh=h)
    .with_columns(pl.col('ds').dt.cast_time_unit(train.schema['ds'].time_unit))  # this is also an issue
    .rename({'y': 'prediction'})
)
plot_series(train, valid.join(y_pred, on=['unique_id', 'ds']))

This outputs the following:

Expected behavior
The seasonal pattern should be repeated, a linear model is easily able to predict this using the correct lags.

Screenshots
I went through the code a bit and I see that the state that is used sets the lag1 as the lag1 at the last timestamp, which is the lag2 when predicting.

I'd expect y__lag_1 to be 0.3 here

Desktop (please complete the following information):

• OS: Linux
• Version: 0.9.5