release-22.1: sql: fix left semi and left anti virtual lookup joins

[yuzefovich]

Backport 1/1 commits from #92713.

/cc @cockroachdb/release

---

This commit fixes the execution of the left semi and left anti virtual
lookup joins. The bug was that we forgot to project away the looked up
columns (coming from the "right" side) which could then lead to wrong
columns being used higher up the tree. The bug was introduced during
22.1 release cycle where we added the optimizer support for generating
plans that could contain left semi and left anti virtual lookup joins.
This commit fixes that issue as well as the output columns of such joins
(I'm not sure whether there is a user facing impact of having incorrect
"output columns").

Additionally, this commit fixes the execution of these virtual lookup
joins to correctly return the input row only once. Previously, for left
anti joins we'd be producing an output row if there was a match (which
is wrong), and for both left semi and left anti we would emit an output
row every time there was a match (but this should be done only once).
(Although I'm not sure whether it is possible for virtual indexes to
result in multiple looked up rows.)

Also, as a minor simplification this commit makes it so that the output
rows are not added into the row container for left semi and left anti
and the container is not instantiated at all.

Fixes: #91012.
Fixes: #88096.

Release note (bug fix): CockroachDB previously could incorrectly
evaluate queries that performed left semi and left anti "virtual lookup"
joins on tables in pg_catalog or information_schema. These join types
can be planned when a subquery is used inside of a filter condition. The
bug was introduced in 20.2.0 and is now fixed.

Release justification: bug fix.

[blathers-crl]

Thanks for opening a backport.

Please check the backport criteria before merging:

• Patches should only be created for serious issues or test-only changes.
• Patches should not break backwards-compatibility.
• Patches should change as little code as possible.
• Patches should not change on-disk formats or node communication protocols.
• Patches should not add new functionality.
• Patches must not add, edit, or otherwise modify cluster versions; or add version gates.

If some of the basic criteria cannot be satisfied, ensure that the exceptional criteria are satisfied within.

• There is a high priority need for the functionality that cannot wait until the next release and is difficult to address in another way.
• The new functionality is additive-only and only runs for clusters which have specifically “opted in” to it (e.g. by a cluster setting).
• New code is protected by a conditional check that is trivial to verify and ensures that it only runs for opt-in clusters.
• The PM and TL on the team that owns the changed code have signed off that the change obeys the above rules.

Add a brief release justification to the body of your PR to justify this backport.

Some other things to consider:

• What did we do to ensure that a user that doesn’t know & care about this backport, has no idea that it happened?
• Will this work in a cluster of mixed patch versions? Did we test that?
• If a user upgrades a patch version, uses this feature, and then downgrades, what happens?

[cockroach-teamcity]

This change is 

[rytaft]

Reviewed 4 of 4 files at r1, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @msirek)