Improve pipeline documentation

[magdapoppins]

Could @DrazenDodik and others interested add existing docs painpoints here? What is bad?

[drazendee]

• [ ] Focuses on the YAML-definition, there are no examples of how to run pipelines over APIs and that it's even possible without a yaml file.

• [ ] Doesn't have an explanation of how node inputs are overridden and where do you read the inputs from

• [ ] What's the default behavior for node override, if the previous step didn't generate an output?

The edges/nodes definitions could use more example and can be a bit confusing. For example, on the edges definitions, what is the difference between node/type/key and how do they map to folders.

Another small thing we noticed today when running pipelines is about how intermediary folders get passed between nodes. If we set a nodes output to output.predictions* and the predictions folder itself contains more folders, only the leaf files get passed to the input of the next node. The documentation is not very explicit on this