Refactor Blueprint Mixins, create ScreenTaskRequired mixin

[JackUrb]

Overview

Creates the first full-fledged Blueprint mixin ScreenTaskRequired, in a similar vein to OnboardingRequired, but with clearer delineation for workflow-specific functionality. This PR also refactors both of these to follow the new BlueprintMixin structure, which lets us standardize how workflows for blueprints can be mixed in. The goal here is to create a task that, for workers that haven't completed any of the task yet, you can put out a test task for which you can more easily validate the results to than a full task. This launches new tasks on the fly, such that you can have validation tasks be separate from the rest of the pool.

Implementation

• Creates the ValidationRequired mixin, which defines behaviors to properly validate and initialize blueprints with this functionality. Defines some logic hooks that Mephisto's ops layer can use to interact with this workflow. Also defines utility methods that are useful in user scripts for those utilizing this workflow.
• Updates TaskRun to ignore special units (for gold, validators, etc) when determining eligible units for workers
• Passes the TaskLauncher along to the Supervisor to be able to handle launching validator units. This smells but I aim to address this kind of over-coupling in the coming refactor of the Supervisor.
• Updates the Supervisor to run validation tasks for new workers on blueprints that have ValidationRequired. Ensures that these are expired when abandoned, to not pass to others.
• Creates a function in the TaskLauncher that creates and launches the special validator units.

Considerations

• This implementation is just getting things working, but really displays the need for a general supervisor refactor. Moving some of the business logic into blueprints and feature mixins will likely help encapsulate things.
• This introduces a specific way to determine if launched tasks are used for special purposes, keyed on the unit_index. I think this is a decent way to go about this, but I wonder if mixins should also have considerations for review scripts? Likely need to update the DataBrowser to handle these better in the future.
• Right now I've dumped the mixins with the Blueprint file, but I think these should go somewhere better. Will break out before merging

Testing

In internal

• Fix automated testing broken by passing task_launcher to supervisor
• write automatic testing for this new blueprint mixin

[codecov-commenter]

Codecov Report

Merging #566 (e4515f5) into main (ef8a831) will increase coverage by 0.87%.
The diff coverage is 80.93%.

@@            Coverage Diff             @@
##             main     #566      +/-   ##
==========================================
+ Coverage   65.45%   66.32%   +0.87%     
==========================================
  Files          79       81       +2     
  Lines        7297     7459     +162     
==========================================
+ Hits         4776     4947     +171     
+ Misses       2521     2512       -9     

Impacted Files	Coverage Δ
...ueprints/static_html_task/static_html_blueprint.py	43.54% <ø> (-0.21%)	⬇️
...ractions/blueprints/mixins/screen_task_required.py	71.59% <71.59%> (ø)
...tractions/blueprints/mixins/onboarding_required.py	74.00% <74.00%> (ø)
mephisto/abstractions/blueprint.py	73.75% <76.92%> (+0.80%)	⬆️
...lueprints/abstract/static_task/static_blueprint.py	41.83% <100.00%> (+0.83%)	⬆️
...sto/abstractions/blueprints/mock/mock_blueprint.py	87.75% <100.00%> (+2.04%)	⬆️
...ns/blueprints/parlai_chat/parlai_chat_blueprint.py	36.98% <100.00%> (+0.43%)	⬆️
mephisto/abstractions/test/architect_tester.py	95.55% <100.00%> (ø)
mephisto/data_model/task_run.py	83.41% <100.00%> (ø)
mephisto/operations/operator.py	57.44% <100.00%> (+0.36%)	⬆️
... and 13 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ef8a831...e4515f5. Read the comment docs.

[pringshia]

Overall looks good! Will re-review pending some of the changes we discussed, e.g. adding max_validations @ the blueprint level, as well as perhaps asserting that this value is necessary to re-emphasize to the user that there is a cost for these tasks

[pringshia]

I have a few copy and terminology related notes, but otherwise this looks good to me.

[pringshia]

Really like this API! ❤️

screening_data_factory could be another possible name here... just throwing out ideas