[ENH] in-sample forecasting with alternative exogeneous `X`, e.g., for scenario analysis

As @ngupta23 pointed out in #3224, forecasters that make use of exogeneous data should support in-sample forecasts with exogeneous data different from the data seen - this is useful and desired, say, in the use case of scenario analysis.

I was under the impression that forecasters support this out-of-the-box, but it seems to depend on the forecaster.

Therefore, we should collect come examples of current failures and see how we can support this best.

Some options:

• tag for forecasters that can do this vs not
• testing this via specific test scenario