Fix ResolveRefs may not resolve a resolvable aggregation with grouping

[jackpan123]

Closes #116781
When the plan first enters the execution of resolveAggregate method, we can observe that c as an UnresolvedAttribute object has a value for the unresolvedMsg attribute which is not null. However, the customMessage attribute is false. But In the potentialCandidatesIfNoMatchesFound method, calling withUnresolvedMessage to regenerate the UnresolvedAttribute object sets the customMessage attribute just based on whether unresolvedMessage != null. This behavior may be a little confusion.

Thus, relying solely on the customMessage attribute of UnresolvedAttribute for some judgments is insufficient and may lead to errors. To address this, I have added additional checks to ensure that the maybeResolveAttribute method and potentialCandidatesIfNoMatchesFound behavior as we expected.

[jackpan123]

@elasticsearchmachine generate changelog

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[wchaparro]

@alex-spies should we take a look again on this one?

[alex-spies]

Thanks a lot @jackpan123 and I'm terribly sorry it took so long to get back to you and review your submission.

Unfortunately, the change proposed here has a pretty large blast radius; while the change is small in terms of changed lines, it changes maybeResolveAttribute. This method is used all over in the reference resolution step, whereas the scope of the bug is limited to very specific edge cases when resolving aggregates.

Running the VerifierTests on this PR shows 4 failures where we return wrong error messages to users, and in one case even the analyzer doesn't terminate anymore. For this reason, I think we should attempt a different approach.

I'm closing this - sorry @jackpan123 .

In the meantime, @fang-xing-esql is working on a more contained fix right now that is limited to the resolution of aggs, only.