ESQL: Fix projection generation when pruning left join

[bpintea]

Currently, in case the join key is (locally) null, the whole join is replaced with an Eval and a Project. The project however contain only reference attributes. This is insufficient, in case the attribute should point to another attribute pointing to a literal null added in the Eval.

This fixes that by allowing the generated Project to contain NamedExpressions, which can then be proper Aliases.
This is done by adjusting the join output calculation to work with the NamedExpressions, instead of just Attributes.

[elasticsearchmachine]

Hi @bpintea, I've created a changelog YAML for you.

[elasticsearchmachine]

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

[astefan]

LGTM
For someone like me (who takes a PR for a spin with a sample query and dissects its internals and results) I would like to also see a concrete query in csv-spec file. You shared one with me offline and I think it wouldn't hurt to also have that one bundled in the PR, as well.
from languages_lookup_non_unique_key | eval language_code = null::integer | lookup join languages_lookup_non_unique_key on language_code

[astefan]

Actually... (sorry about that), I had an idea for a test and it turns out we don't have a test for that and with this PR it has a wrong behavior:
from languages_lookup_non_unique_key | keep country, language_name | eval language_code = null::integer | inline stats max(language_code) by language_code

Before this PR, the result is:

        country         | language_name |max(language_code)| language_code 
------------------------+---------------+------------------+---------------
Canada                  |English        |null              |null           
null                    |English        |null              |null           
United Kingdom          |null           |null              |null           
...

with this PR, the order of the columns changes (and it's wrong):

        country         | language_name | language_code |max(language_code)
------------------------+---------------+---------------+------------------
Canada                  |English        |null           |null              
null                    |English        |null           |null              
United Kingdom          |null           |null           |null              
...

[bpintea]

I had an idea for a test and it turns out we don't have a test for that and with this PR it has a wrong behavior

Indeed, due to the updated method overriding, the InlineJoin ended up using the wrong method. Thanks for catching it, @astefan! I've updated that and added in the test.

[bpintea]

These changes make insure that the country field is part of the languages_lookup_non_unique_key schema. Otherwise the CSV tests generate an index resolution without it, which leads to failed queries when using the index in FROM.
The index was previously using similar languages schema, which lacks the fields, but which was added in dynamically by the data. Also, the index had only been used in LOOKUP JOIN before.
The change then also leads to updated order of the MV value ["Germany", "Austria"]

[astefan]

LGTM

[bpintea]

Thanks, Andrei!

[elasticsearchmachine]

💔 Backport failed

Status	Branch	Result
❌	8.19	Commit could not be cherrypicked due to conflicts
✅	9.2

You can use sqren/backport to manually backport by running backport --upstream elastic/elasticsearch --pr 135446