feat(workflow): workflow revert command

[mohammad-alisafaee]

Description

• renku workflow revert command. There are some deviations from the story definition:

  1. --soft is renamed to --metadata-only
  2. --plans is renamed to --plan
  3. The story say that generations should be changed if the activity is the newest execution and has some downstream activities. The implementation always reverts the generations to an earlier version (if any).
  4. ActivityGateway.get_all_activities won't return deleted activities by default.
  5. When an activity has downstream activities, it doesn't prompt to ask if it should revert the activity. Instead, it always fails and asks users to pass --force in such cases.

• When deleting a plan, it creates a derivative plan and marks it as deleted.

Fixes #2342
Fixes #2969

NOTE

We don't detect files that are used in datasets and might still delete them if they are not used in a workflow.

[Panaetius]

Amazing how quick this was done, great work!

We should also have someone from ORDES review this (with some instructions) once we settled on a stable state.

[Panaetius]

This is a bit of a break from how we do deletions on datasets, where we just add a new derivation that has invalidated_at set.

I wonder if we should bring these in line. Also not sure how it affects KG if they get modified Plan nodes

[mohammad-alisafaee]

I already mark all activities in the activity chain as deleted. If that's an issue for KG then we shouldn't do that and #2954 cannot be done. I'll talk to KG about this.

[Panaetius]

when I added renku workflow remove, the idea way that a Plan chain ends in a node that is deleted, it means all versions are deleted. I think the filtering out of deleted plans in other commands was don't in parallel in a separate PR and only the nodes that are invalidated are ignored (but not their parents). I think the latter is wrong and we should treat it the same as datasets (an invalidation invalidates the whole chain).

For activities it's fine since they can't be edited (just removed).

[mohammad-alisafaee]

I've talked to Kuba about this and it's OK to set a property that didn't have a value before (prov:wasInvalidatedAt in this case). This may affect project's fork if deletion was done after the fork (all forked projects see the plan as deleted) but they will fix this in future by name-spacing forks since this happens for all other properties as well.

[Panaetius]

One minor change, otherwise looks great 👍