Better decoupling from flaml

[baggiponte]

We should be installing flaml[automl] instead. Since it depends on pandas and some other libraries, it should be made a separate dependency group. However this requires quite a lot of refactoring:

1. Not straightforward:

   functime/functime/forecasting/_regressors.py

   Line 150 in 092743a

   class FLAMLRegressor:

   1. Requires moving this forecaster somewhere else:

      functime/functime/forecasting/lightgbm.py

      Line 118 in 092743a

      class flaml_lightgbm(Forecaster):

2. Easy:

   functime/functime/forecasting/automl.py

   Line 8 in 092743a

   from flaml import tune