Ensure MESSAGE.enforce() runs for MESSAGE_MACRO

[khaeru]

Closes #591 by ensuring that the MESSAGE.enforce() method, ensuring consistency of internal sets and parameters, is called when a scenario is solved using the MESSAGE_MACRO class.

Note that this demonstrates the workflow described at #603 (comment):

1. Author opens a PR with some commits. Status in our tracker is "In progress".
2. Identifying that their PR touches the GAMS implementation involves the nightly tests, author pushes a commit uncommenting these lines:

   message_ix/.github/workflows/nightly.yaml

   Lines 6 to 8 in b572cce

   	# Uncomment these two lines for debugging, but leave them commented on 'main'
   	# pull_request:
   	# branches: [ main ]

   with a message like TEMPORARY Enable nightly workflow for PR.
3. Author develops their PR, pushing commits, noting and ensuring that the "pytest", "lint" and "nightly" checks pass.
4. Author requests a review; status is "In review".
5. Reviewer does the review, possibly requesting changes.
6. Review and author interact until reviewer is satisfied.
7. Reviewer approves.
8. Author or maintainer performs git rebase -i on the branch to drop the "TEMPORARY" commit from step 2.
9. Merge.

How to review

• Read the diff and note that the CI checks all pass.
• Run specific scenarios as described at Address failing nightly workflow  #591 (comment) and ensure the behaviour remains the same as prior to the merge of Resolve the issue of zero capacity factor #561

PR checklist

• Continuous integration checks all ✅
• Add or expand tests; coverage checks both ✅
• Add, expand, or update documentation. N/A, fix bug to restore intended/documented behaviour
• Update release notes.

[khaeru]

Prior to implementing a fix, the "nightly" CI workflow, temporarily enabled, fails: https://github.com/iiasa/message_ix/runs/6593114663#step:9:64 —as seen on the scheduled runs on main.

[codecov]

Codecov Report

Merging #613 (e605bfe) into main (2ae918e) will decrease coverage by 0.0%.
The diff coverage is n/a.

@@           Coverage Diff           @@
##            main    #613     +/-   ##
=======================================
- Coverage   93.0%   93.0%   -0.1%     
=======================================
  Files         43      43             
  Lines       3375    3374      -1     
=======================================
- Hits        3139    3138      -1     
  Misses       236     236             

Impacted Files	Coverage Δ
message_ix/message_ix/models.py	100.0% <0.0%> (ø)

[khaeru]

Now with the fix, the "nightly" CI workflow succeeds: https://github.com/iiasa/message_ix/runs/6593565149?check_suite_focus=true#step:9:42

(With this comment, I am making this fact visible to whoever reviews the PR.)

[khaeru]

I've restarted one flaky test (failure is the in the R notebook, so unrelated), but otherwise this is complete. So now I'll request a review, with both the failing and passing runs visible.

[gidden]

LGTM - elegant solution to reduce code duplication, but does introduce the diamond problem. For now let's just note that here. I highly doubt we will ever run into it given we do not touch these classes often. A pythonic resolution to this for future searchers can be found here (if it ever becomes a problem).

Happy to merge as is - should we ignore the tutorial failure?

[gidden]

all tests pass, merging

[khaeru]

@gidden —thanks for the review, but the merge was a little premature.

Like the comments/description said, I was trying to demonstrate the simple workflow for running the nightly tests on a PR branch. Now we've skipped from step (7) to (9), missing (8), so changes from the "TEMPORARY" commit now appear on main.

I'll revert and see if I can fix.

[khaeru]

Done via #614.