Cannot compute predictions on yearly dataset spanning 200 years

Repro

```python
import prophet
import pandas as pd
import pytest

df = pd.read_csv("/Users/freddy.boulton/Downloads/frequency_A.csv")
df = df.rename(columns={"Date": 'ds', 'Target': 'y'})
df['ds'] = pd.to_datetime(df['ds']).dt.tz_localize(None)
regressor = prophet.Prophet()

df_train, df_test = df.iloc[:240], df.iloc[240:]
df_test.pop('y')

regressor.fit(df_train)

with pytest.raises(OverflowError, match="Overflow in int64 addition"):
regressor.predict(df_test)
```

The error is coming from `df['t'] = (df['ds'] - self.start) / self.t_scale` which does not support long time-spans based on this [thread](https://stackoverflow.com/questions/56720783/how-to-fix-overflowerror-overflow-in-int64-addition)