(Closes #2125) Fix bugs in SymbolTable deep copy and InlineTrans when array bounds reference formal arg

[arporter]

Fixes bugs in SymbolTable deep copy for for symbol initial values and shape specification. Also fixes a bug in InlineTrans when the lower bound of an array formal argument is specified by a reference to another formal argument.
Finally, adds the force option to InlineTrans to permit it to ignore CodeBlocks within the target routine.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.85%. Comparing base (029f69f) to head (565c2c3).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2633      +/-   ##
==========================================
- Coverage   99.86%   99.85%   -0.02%     
==========================================
  Files         352      352              
  Lines       48512    47308    -1204     
==========================================
- Hits        48447    47239    -1208     
- Misses         65       69       +4     

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

[arporter]

This is ready for a first look now. One for @sergisiso, @hiker or @LonelyCat124. I've triggered the integration tests.

[arporter]

(I am wondering whether having the "force" option to InlineTrans mean "ignore CodeBlocks" is a good idea: perhaps it should be more specific such as "permit-codeblocks"?)

[sergisiso]

@arporter I agree this needs to be done, as now we carelessly leave symbols pointing behind. But there are more symbols that the ones updated by this PR and the behaviour would be inconsistent if we update some but not others. Let me know if you want this PR sooner to fix a specific bug, but otherwise I suggest also updating all symbols inside symbols/datatypes.

[sergisiso]

Fantastic work @arporter. There is final documentation suggestions but then this can go.

[sergisiso]

Also check if the sphinx-link-check issue is relevant and launch an integration test since I see none of the last ones were complete.

[arporter]

Link check failure seems to have sorted itself out. I've triggered the integration tests again. Ready for another look, CI-permitting.

[arporter]

LFRic 'everything-everywhere-all-at-once' integration test failed due to hitting max. recursion depth.
Problem occurs in backend when processing (inlined?) kernel mm_diagonal_kernel_code_r_double for reconstruct_w3_field_alg_mod_psy.

[arporter]

The problem was that in Reference.replace_symbols_using() I had added a call to replace_symbols_using on the Symbol itself. This then hit an infinite recursion if the Symbol had an initial value defined by a PSyIR expression. Since it is the responsibility of the SymbolTable to update its symbols, I've simply removed the offending call and updated a failing test.

[arporter]

Integration tests were green and coverage is all OK. Really ready for review...

[sergisiso]

@arporter All comments have been addressed, the documentation has been updated explaining the new internal details and it builds correctly. The integration passes (and the system was noisy but looking at the best bars on multiple runs it seems the performance is unaffected). This is ready to merge.