Integration of forecasting algorithms from Nixtla

[sunng87]

What problem does the new feature solve?

The ultimate goal is to add built-in time series forecasting support, with an extensible plug-in architecture.

What does the feature do?

We can start from those open source forecasting engines from nixtla. Including:

• https://github.com/Nixtla/neuralforecast
• https://github.com/Nixtla/statsforecast

We can already evaluate these features by connecting greptimedb to mindsdb, in which nixtla forecast engines are built-in.

Implementation challenges

• It can be a challenge to build python code into the database, especially with pytorch
• Provide a SQL extension for forecasting
  • allows selecting algorithms/engines from a range of choices
  • may require additional step to build the model