cleanups: todo after implementation of pipeline file

[skshetry]

Tracking things to cleanup after #3676 is merged, in addition to suggestions in the PR:

• Remove tags (will make life simpler) tag: getting rid of it #3699
• Split Repo.collect() for pipelines usage and stage gathering:
  Data related commands don't care about orderings of pipeline stages, whereas for pipelines,
  things are more complex than collect() can handle.
  This will also remove duplicated warnings currently on reproduce.
  Workaround on tag: getting rid of it #3699 for now
• Why is there schema/serialization/deserialization in 3 places? Yeah, true.
• Parallel logics of loading/transforming/dumping of Output, Dependency and Stages for single stage and pipeline stages (Requires further investigation)
• StageLoader is too dynamic.
• Messed with OutputDuplicationsError: Make sure it's safe to remove current stage from error message.
• (Feature Request) Some way to remove stage from both pipelines.lock and pipelines.yaml? (Implement removing of the stage #3881)
• Renaming utils/stage.py and functions inside it. (refactor: rename utils.stage to utils.yaml #3923)
• Make dump() ordered (Only for Python==3.5).
• StageParams and OutputParams

Of course, we need lots of tests and UI issues fixes, but this meta issue is only regarding code cleanups.

[jorgeorpinel]

I find the long name of options --metrics and --plots misleading, since they only accept one metrics/plot file. What about just --metric and --plot? Cc @dmpetrov

[efiop]

@jorgeorpinel Those are just consistent with the rest of the options: --outs/--outs-no-cache/etc. We might want to introduce an alias or support commas like we do in --params. But for now could leave as is.