YAML based config

[avishniakov]

I reworked the template to use the YAML configuration instead of the Python file. This makes this template reusable for a broader count of use cases and makes it aligned with general ZenML configuration best practices.

• Configuration for HP tuning and other uses is now passed as extra config via YAML
• Search space objects support lists or ranges. For range a dictionary of {"range":{"start":1,"end":10,"step":2}} is used.
• Artifacts folder and inner objects are no longer needed
• New get_model_from_config step to reconstruct ClassifierMixin from string-based configurations

Depends on zenml-io/zenml#1876

[fa9r]

Not really familiar with config best practices so I'm tagging @bcdurak instead who worked on this before I believe

[avishniakov]

MacOS tests are broken, to be followed up here

[strickvl]

Nothing to do with this, but while I can see that people might prefer this kind of config for simple configurations, but I find this a decent bit less clear than having a Python-based config.

[avishniakov]

Yeah, I also prefer Python, but it is breaking caching logic currently and might not lead to a new pipeline version generated in ZenML on config update.