[ENH] building a systematic test suite for models

[fkiraly]

We should build a systematic test suite for models.

Currently, there are tests per model.

A "test suite" means something like TestAllForecasters in sktime, where we loop over forecasters, but with the contract of pytorch-forecasting.

The test suite should cover basic usage vignettes, including those shown in the tutorials.

Architecturally, I suggest using scikit-base as a soft dependency (testing depset only), to index and collect all forecasters, and to build the test framework.

While this is probably not a good first issue, we should collect code snippets that should work for all foercaster classes, for inclusion in the test suite.

[PranavBhatP]

@fkiraly, is there any pending work on this? What could I probably focus on, going forward with this issue?

[fkiraly]

This is a difficult task - I have started a draft pull request here:
#1780

The next step would be to complete this for the current API.

The missing pieces are:

• identify points of variation across current tests and map them onto a suite test
• move tests for more model classes to the test suite

Let me know if you are planning to work on this.