Update correlation in Apply node

[kokosing]

Update correlation in Apply node

All the column references of subquery plan which are not resolved (do
not come from subquery relations) are collected. SubqueryPlanner check
if it is able to rewrite such references to symbol produced by its
relation, if it is able then add replace column reference with that
symbol and that this symbol to the correlation list of Apply node. If
SubqueryPlanner is not able to rewrite that column reference, then it is
assumed the outer SubqueryPlanner execution will do that. If there is no
outer SubqueryPlanner then SemanticException should be thrown.

[kokosing]

Supersedes #5302.
Please notice that this pull request is based on #5686

[ArturGajowy]

return resolvedField.filter(ResolvedField::isLocal).map(field -> outputSymbols.get(field.getFieldIndex()))

[kokosing]

done

[haozhun]

I did not read this pull request. But I would like to add a comment because of a production issue we've seen before. When Teradata brought non-equi conjunct support to outer join in Presto (thank you again), we removed the sanity check that fails any query that contains such conjuncts (and its tests) before the feature were linked together to execute such query correctly. As a result, for more than a month, the Presto release version produces incorrect result for such queries.

Please make sure any query with correlated subqueries continues to fail loudly until such query produces correct result.

[kokosing]

@haozhun That is a general true. None of the features should be enabled until it works correctly. I am trying to test it as mush as I can, nevertheless the bugs may still exists. That is why code review is so much important, that I could benefit from your experience and possibly broaden the test coverage to avoid that situation you (FB) experienced before.

[kokosing]

@haozhun Maybe we could display a notice to the user about possible instability of new features, that given feature was recently added and may still have bugs or instabilities. Such notice could be displayed through couple of Presto releases. What do you think about such approach, would it help if you had something like that for non-equi conjunct support to outer join back then?

[martint]

This should probably in a separate commit.

Also, can you remind me of the rationale for this? I remember discussing about this, but I can't find the original comments.

[kokosing]

Fixed. I added a commit message which explains the motivation.

[kokosing]

@martint Rebased. Comments applied. Superseded by #5943