ESQL: PushdownRegexExtract doesn't take into account shadow attributes

[costin]

Description

Grok and Dissect produce attributes which can clash with existing fields creating new attributes (with the same name).
When reordering this through the tree however, the shadow attributes are not considered leading to inconsistencies in the tree:

// notice the two emp_no attributes
 [test] Rule optimizer.LogicalPlanOptimizer$PushDownRegexExtract applied
...
  \_Dissect[full_name{r}#5,Parser[pattern=%{emp_no} %{b}, appendSeparator=, parser=org.elasticsearch.dissect.DissectParser@ !   \_OrderBy[[Order[emp_no{f}#15,ASC,LAST]]]
15e5e308],[emp_no{r}#6, b{r}#7]]                                                                                            !     \_Dissect[full_name{r}#5,Parser[pattern=%{emp_no} %{b}, appendSeparator=, parser=org.elasticsearch.dissect.DissectParser@
    \_OrderBy[[Order[emp_no{f}#15,ASC,LAST]]]                                                                               ! 15e5e308],[emp_no{r}#6, b{r}#7]]

[elasticsearchmachine]

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

[alex-spies]

PushDownEval has the exact same problem.

I have reproducers for both that result in NPEs if run as CSV tests:

//eval
FROM employees
| SORT emp_no ASC
| eval emp_no = -emp_no
| LIMIT 3
| KEEP emp_no

//dissect
FROM employees
| MV_EXPAND job_positions
| SORT emp_no ASC
| DISSECT job_positions "%{emp_no} %{} %{}"
| LIMIT 1
;