Get started: better dataset, dvclive, more plots

[shcheklein]

Based on initial work done by @daavoo in this PR - https://github.com/iterative/example-repos-dev/pull/102

Depends on https://github.com/iterative/dataset-registry/pull/28. A few things will be improved here when that PR lands.

Corresponding repo to play with: https://studio.iterative.ai/team/Iterative/views/example-get-started-zde16i6c4g

(a few lines in the PR will be uncommented / reverted before actual merge, after testing)

It changes example get started:

• To use a clean dataset (thus 2.5x smaller)
• Less number of features (3000 -> 100/200). Since dataset is better it's now enough - should be way better memory footprint (prepared matrix is 4x smaller)
• More plots - confusion matrix + forest analysis (most useful feature)
• CML report includes confusion matrix
• Two branches - more data, hyperparams tuning

[shcheklein]

Some things will be uncommented and reverted after data registry update lands.