checkpoints/live: move execution env to stage

[pared]

• ❗ I have followed the Contributing to DVC checklist.

• 📖 If this PR requires documentation updates, I have created a separate PR (or issue, at least) in dvc.org and linked it here.

Thank you for the contribution - we'll try to review it as soon as possible. 🙏

This PR aims to make Output source of the execution environment for Stage.run. It seems that outputs are the cause of env.
Related: #4902 (comment)

[pmrowla]

This looks ok to me with the way checkpoints and dvclive are currently defined, but there has been some discussion on whether or not checkpoint should just be a stage level flag in the future, rather than an output level flag. Not sure if the same consideration would apply to dvclive?

[pared]

@pmrowla
Well, as of now, there can be only one dvclive defined. And env is resource used mainly by stage, so it does make sense to me to define it on stage level. I can move env out of out into stage. How about that?

[pmrowla]

@pmrowla
Well, as of now, there can be only one dvclive defined. And env is resource used mainly by stage, so it does make sense to me to define it on stage level. I can move env out of out into stage. How about that?

I think having it in stage makes more sense 👍

[pared]

@pmrowla moved env to stage

[efiop]

@pared Looks like PR and commit titles need an update too?

[pmrowla]

This isn't because of your changes, but looking at this now, I think we should be setting DVC_ROOT all the time in the calling env() function. (DVC_CHECKPOINT should still only be set inside this if case)

For experiments that are run in temp-dirs (and not only checkpoint experiments), it will always be useful for user's code to be able to see where it is being run, since they may need to work with relpaths from that directory (instead of relpaths inside their git/dvc repo workspace). And outside of the experiment context, it won't hurt us to set DVC_ROOT for regular repro runs as well.

[pared]

Good point!