(Closes #2854) exclude char assignments from ACC KERNELS regions

[arporter]

Small PR that picks up the functionality already in ArrayAssignment2LoopsTrans and copies it into ACCKernelsTrans.

[arporter]

I'm not entirely happy with the code duplication here but this PR is for OpenACC functionality and the existing check is in the conversion of array assignments to loops so there's not much common ground between them.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.89%. Comparing base (e2d88e9) to head (3c92b65).
Report is 10 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2857   +/-   ##
=======================================
  Coverage   99.89%   99.89%           
=======================================
  Files         359      359           
  Lines       50981    50995   +14     
=======================================
+ Hits        50926    50940   +14     
  Misses         55       55           

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

[arporter]

I've solved the duplication by making the validation a separate classmethod in ArrayAssignment2LoopsTrans. Coverage is 100%. I'll set the integration tests running.

[arporter]

Integration test for NEMOv4 failed but only because of a small bug where we failed to lower-case the name of a structure component when getting its datatype. I'll re-run that job.

[arporter]

Integration tests now green. Ready for review from anyone :-)

[arporter]

I changed my mind - it turns out my 'fix' revealed general case-sensitivity problems in our storage of components of a StructureType. I've fixed these now and extended the tests slightly.

[arporter]

Ready for review now. Integration tests are finally happy :-)

[LonelyCat124]

@arporter Couple of small changes in docs, and a couple of queries I'm just unsure about.

[arporter]

Imaran has found that this PR doesn't fix e.g. obs_surf_def.f90 because that has a character string as a member of a derived type and we haven't resolved the derived type so don't know that it's a character string. There are a few options:

1. always attempt to resolve the type in such a case
2. refuse to allow unresolved types inside ACC regions
3. exclude the file from PSyclone processing

I think we want 2. to be the default and 1. to be an option. I thought @sergisiso had an Issue about this (being much more defensive about what we offload) but I can'f find it?

[arporter]

Although this doesn't fix all our (OpenACC) problems, it is a step forward so I've opted to keep this PR small and self-contained and not attempt to fix anything else. Ready for another look now. I'll fire off the integration tests as master has moved on a lot since I last did that.

[LonelyCat124]

Integration tests work - am going to have one more local pass before marking this ready to merge.