ESQL: Remove obsolete code path for aggs

[alex-spies]

Essentially a forward-port of #117832

Since #104958, AggregateExec cannot have aliases in its aggregates - because any such aliases are turned into Evals.

But before that, a STATS command like

...
| rename languages as l
stats c = count(e) by l

would result in an AggregateExec like

AggregateExec[[l{r}#1],[COUNT(emp_no{f}#4,true[BOOLEAN]) AS c, languages{f}#5 AS l],INITIAL,[l{r}#1, count{r}#10, seen{r}#11],nu = AggregateExec[[l{r}#1],[COUNT(emp_no{f}#4,true[BOOLEAN]) AS c, languages{f}#5 AS l],INITIAL,[l{r}#1, count{r}#10, seen{r}#11],nu
ll]  

where the renaming languages{f}#5 AS l is pushed into the aggregates and the groupings refer to it via the reference attribute [l{r}#1].

AbstractPhysicalOperationProviders.groupingPhysicalOperation still addresses this remnant of times long past by checking whether AggregateExec has any Aliases in its aggregates, and adjusting the aggregation's layout so it points to the source attribute languages{f}#5, not the renamed attribute l{r}#1 (because the renamed attribute is not present in the source layout!)

Let's get rid of this dead code since it is confusing.

[elasticsearchmachine]

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

[alex-spies]

Reverting to draft: some union type tests are failing, so it seems this code path is still used.