fix: make column.unnest().topk() work

[NickCrews]

Also simplifies the implementation of Column.topk() and
makes it re-use the implementation of Table.topk()

This does NOT fix column.unnest().topk(by=<anything besides None>). This is a rare case so I decided to just punt on implementing it

This also factors out the common marks for when arrays, structs, maps, json are not supported by a backend. I wanted to reuse that mark, and I bet there are many existing tests that could benefit from this utilility.

[NickCrews]

I split out the test refactoring into #11315, maybe we merge that and then I revade this to only include the behavior changes

[NickCrews]

@cpcloud This is now a lot simpler now that the test refactor is merged separately, and hopefully should be a quick review.

[cpcloud]

Seems to be failing more than half of CI checks.

[cpcloud]

Given that grouping by UNNEST works in postgres, I'd much rather get this implemented upstream in DuckDB than hack around it here.

[cpcloud]

Opened a feature request in DuckDB for adding support for grouping by unnests: duckdb/duckdb#17884

[cpcloud]

I wonder if there's some DuckDB specific thing we can do here that would take any query with SELECT UNNEST(x) FROM t and rewrite it as SELECT * REPLACE (UNNEST(x) AS x) FROM t.

[cpcloud]

The hairy part is of course that UNNEST can be buried in an arbitrarily deep expression such as 1 * 2 * LOG(UNNEST(x))