ML Pipelines: recommend push=false for intermediate outputs

[jorgeorpinel]

Echoing iterative/dvc#8581 (comment):

Should we plan to recommend this a lot in Data Pipeline docs? Specifically for intermediate pipeline outputs. Assuming the happy path out there is to push only raw data and likely final ML model files (everything else may be best to dvc repro when needed).

If we don't at least emphasize the possibility, users may realize too late they have pushed a bunch of intermediate output versions and they are pretty difficult to clean up with dvc gc (support example).

Originally posted by @jorgeorpinel in #4142 (comment)

See push field in https://dvc.org/doc/user-guide/project-structure/dvcyaml-files#output-subfields and https://dvc.org/doc/user-guide/project-structure/dvc-files#output-entries.

[jorgeorpinel]

let's brainstorm more on it (#4142 (comment))

Cc @dberenbaum hoping this issue is helpful for that purpose. Added task in #2883 for now too.