Add backward compatibility for supported version of scikit-learn

PRs #957 and #959, updated `dask-ml` to the latest 2 releases of `scikit-learn`. We now need to decide on the minimum versions officially supported in `dask-ml`.

From @TomAugspurger:

> In general, my policy was to match the lowest-supported version of scikit-learn, but to pretty aggressively bump that minimum supported version when it became at all challenging to support both versions. If people are able to upgrade their version of scikit-learn to the latest, then they ought to be able to upgrade dask-ml as well (and vice-versa).

From the `scikit-learn` [docs](https://scikit-learn.org/stable/install.html#install-official-release):

> **Warning** Scikit-learn 0.20 was the last version to support Python 2.7 and Python 3.4. Scikit-learn 0.21 supported Python 3.5-3.7. Scikit-learn 0.22 supported Python 3.5-3.8. Scikit-learn 0.23 - 0.24 require Python 3.6 or newer. Scikit-learn 1.0 supported Python 3.7-3.10. Scikit-learn 1.1 and later requires Python 3.8 or newer.

In #957, `dask-ml` dropped 3.7 and added 3.10. `dask-ml` could support `scikit-learn` versions 1.1 or newer, matching the Python versions supported by both `dask-ml` and `scikit-learn`.