Make it easy to re-run an existing pipeline run

This is an evolution of the "meta" file idea (#192 (closed)). Now that we have a "pipeline run" almost ready I think we should replace the meta file support in the current runtime implementation with a way to run a pipeline based on existing pipeline run file. This can support everything a meta file was trying to achieve, but also others. And it would be based on existing standardized file without having to implement a new file definition.

Some considerations:

• YAML pipeline run files can contain multiple pipeline runs, so we can allow storing and reproducing multiple phases of a pipeline run (train, test).
• These files will mostly have to be self-contained except for datasets, problems, and pipelines.
• We should make it easy both saving pipeline runs to a file and then running from it, while also checking the results.