Use more realistic dataset in docs

[aguschin]

Iris dataset used in GS creates a feeling that MLEM is not ready to any serious tasks. Suggestion: use more realistic dataset.

@mnrozhkov suggested to use a dataset from this repo https://github.com/iterative/demo-bank-customer-churn

• Update GS
• Update and re-generate example-mlem-get-started repo with https://github.com/iterative/example-repos-dev, replacing iris with the above dataset
• Update other parts of docs where we reference models/data in example-mlem-get-started (worth completing Test snippets (WIP) #195 beforehand?)

[omesser]

Option: https://www.kaggle.com/competitions/paddy-disease-classification