[feat] Expose @final decorator in the class body

[jjotero]

One now can do

class MyTest(rfm.RegressionTest):
    @final
    def foo(self):
        ...

instead of having to use the @rfm.final syntax.

Other changes:

• Renamed _rfm_special_test to _rfm_override_final. This flag is only used in the metaclass to allow the override of final methods, so this name is more descriptive. Also, the "special" naming of it should just be down to the RegressionTest class.
• The unit test of this functionality is now in unit tests/tests_meta.py instead of the file with the loader tests.
• The run stage of the RunOnlyRegressionTest was calling super().run.__wrapped__(self). I've replaced this line for super.run() since now the final decorator does not create a functors.wrap wrapper.

Closes #2001

[codecov-commenter]

Codecov Report

Merging #2058 (044c6be) into master (34ee3d0) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #2058      +/-   ##
==========================================
- Coverage   86.75%   86.75%   -0.01%     
==========================================
  Files          52       52              
  Lines        9235     9240       +5     
==========================================
+ Hits         8012     8016       +4     
- Misses       1223     1224       +1     

Impacted Files	Coverage Δ
reframe/core/meta.py	98.98% <100.00%> (+0.02%)	⬆️
reframe/core/pipeline.py	91.51% <100.00%> (-0.17%)	⬇️

---

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 34ee3d0...044c6be. Read the comment docs.

[jjotero]

@jenkins-cscs retry daint

[vkarak]

Lgtm. I have just a minor comment.

[victorusu]

@jjotero and @vkarak, I cannot understand why we have a way to override the @final.
If it is final, then it should be final, no?
I would vote for not having @final at all, if one can bypass it in a way that is supported by the framework...
What do you think?
Am I being too radical?

[jjotero]

@victorusu This is needed by the RunOnlyRegressionTest and CompileOnlyRegressionTest classes. The pipeline methods in the base RegressionTest are marked as final to ensure that users do not accidentally override the methods and break the pipeline, but a "master override" method is needed to specialise these two derived classes.

[victorusu]

lgtm

[vkarak]

The only thing we don't check in the unit tests is that a deprecation warning is thrown in case of the old syntax. I think that's why "coverage" is not very happy.

[vkarak]

@jenkins-cscs retry daint