(closes #3188) Improve automatic array privatisation validation

[sergisiso]

This fixes two problems:

• The escapes_scope did not consider that if an array is written to a single element after the loop, the rest of the indices are still potentially escaping the scope.
• The privatisation validation did not consider that a different iteration of the loop may read a value of the array set by another thread, so it should not allow to privatise arrays that may read any position in the array of interest first (as this may not have the right value even if it is firstprivate).

I also made the NEMOv5 gcc tests check full numerical reproducibility, which needed 5 files excluded.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.90%. Comparing base (273a46a) to head (2f41b6f).
⚠️ Report is 13 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #3214   +/-   ##
=======================================
  Coverage   99.90%   99.90%           
=======================================
  Files         376      376           
  Lines       53194    53225   +31     
=======================================
+ Hits        53142    53173   +31     
  Misses         52       52           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sergisiso]

@arporter This is ready for review.

[arporter]

Looks good, thanks for tracking down and fixing these fairly subtle issues. Just some minor tidying required. Your update to the message about adding symbols to "explicitly_private_symbols" made me wonder whether we have this documented anywhere but we don't seem to. Really, it would go in a "how to" section or a tutorial but I don't think we currently have anything like that?

[sergisiso]

@arporter This is ready for another look.

"explicitly_private_symbols" made me wonder whether we have this documented anywhere

Although the current implementation of "explictly_private_symbols" is useful in NEMO, people requested it for LFRic_atm components but was not applicable because they use region directives, not loop directives. So I will attempt to refactor/rethink it before documenting it. But I mentioned this point in #3216

[arporter]

Thanks Sergi, all looks good now.
I see the integration tests were fine.
Will proceed to merge.