[docs] Forecasting with multiple time series with missing values

[ngupta23]

Description

The missing value tutorial shows how to handle missing values when only 1 time series is present in the data, but no guidance is given on what to do when there are multiple time series in the data with missing values. In such cases, individual time series can have different start and end dates.

Having a tutorial (or extending the original one) to show how to handle these scenarios would be helpful for users.

Some decisions that the user can take

1. Should the missing time stamps be added from the global min date in the data or on a per series basis (earliest time stamp per series)
2. Should the missing time stamps be added until the global max date in the data or on a per series basis (latest time stamp per series)

The fill-gaps utility can be utilized for this purpose but it would be great to highlight the various arguments that the user can use in case of multiple time series.

Link

https://docs.nixtla.io/docs/tutorials-missing_values