ESQL: Consistency in non-json responses for profile, _clusters, etc

[nik9000]

Description

JSON, YAML, CBOR, and SMILE are the "standard" ESQL's responses formats. "Everything" work for them. We also support txt, csv, and tsv AND are experimenting with arrow as response formats. "Everything" is not supported in those right now. Right now this all feels kind of haphazard. We need guidelines. A story. Or guiding principals. Something short we can hold in our head and explain in the docs.

Examples:

• The include_ccs_metadata and columnar options throw an error in txt/csv/tsv/arrow.
• txt/csv/tsv/arrow drop the profile response on the floor right now. This proposes failing it like the examples above.

Fun:

• columnar can't work in arrow - it's always columnar. That's what it is to speak arrow.
• We expect people to integrate applications with arrow, so we need include_ccs_metadata to work. The way we've rigged up the arrow format at the moment it only supports streams of pages. I think.
• I'm not sure if we expect applications to consume txt/csv/tsv - They are useful - txt is very human readable and csv/tsv you can dump into a spreadsheet super easily.

[elasticsearchmachine]

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

[nik9000]

It's tempting to say "everything should work everywhere" but I worry we might end up spending a ton of time on txt support for some feature that's deeply about application integration. I dunno. We need to think and make a consistency.