fix(prqlc): improve requirement logic

[Fanaen]

Fixes #5354

New result:

SELECT
  DISTINCT ON (
    EXTRACT(
      year
      from
        hire_date
    ),
    CONCAT(
      'Year ',
      EXTRACT(
        year
        from
          hire_date
      )
    )
  ) CONCAT(
    'Year ',
    EXTRACT(
      year
      from
        hire_date
    )
  ) AS year_label
FROM
  employees

---

About the refactoring

get_requirement is quite logic-heavy.

It made sense to separate column selection logic from complexities/SELECT status logic when there the case were simpler. The special cases with multiple complexities now outnumber general cases.

So I added Requirements: a newtype for Vec<Requirement>. It adds Complexity::Plain and select: false by default (the common case) and a few convenience methods to improve readbility.

It's a separate commit in case we want to split the PR.

[max-sixty]

thanks @Fanaen !

🚧 I'm not sure about testing strategy for this. Input is quite messy, not necessarily something we want in integration tests.

could we add the case as a simple inline test in prql/prqlc/prqlc/tests/integration/sql.rs? that makes it easy to understand the change when the code is open locally. if we can reduce the size of the example a bit then great, but not worth spending much time on making it totally minimal :)

[Fanaen]

Thanks @max-sixty! I added the inline test as suggested. The PR is ready for review!

[Fanaen]

I've found that c1d7069, which I added in an attempt to get nicer SQL results, sometimes removes necessary columns for instance when referenced in a filter.

As it's not necessary for this fix, I've removed it.

[max-sixty]

great! thank you @Fanaen

(for transparency: I tried this out but haven't thought through each line of code, so if we need to iterate more we can do so, thank you for your recent contributions!)

[vanillajonathan]

Perhaps this ought to be mentioned in the CHANGELOG?

[max-sixty]

fine to add things to CHANGELOG but no strong requirement!