Remove unnecessary join in RelationFilter

[KhooHaoYit]

Fixes prisma/prisma#18343
Mitigates prisma/prisma#16450

This PR aims to improve relation filter by removing unnecessary join

How this is achieved is when a model contains fields that can reference to the related table, we can use that fields instead of getting the values from the related table and joining the table
If the fields are stored in the other model, join table would still be performed

There might be performance regression where the related table might return the same identifier multiple time (Many to one) but if that is a concerns we can fall back to join table

Before patch

SELECT "public"."User"."id",
  "public"."User"."email",
  "public"."User"."name"
FROM "public"."User"
WHERE ("public"."User"."id") IN (
    SELECT "t0"."id"
    FROM "public"."User" AS "t0"
      INNER JOIN "public"."Post" AS "j0" ON ("j0"."authorId") = ("t0"."id")
    WHERE (
        "j0"."title"::text LIKE 'a'
        AND "j0"."content"::text LIKE 'b'
        AND "t0"."id" IS NOT NULL
      )
  )

After patch

SELECT "public"."User"."id",
  "public"."User"."email",
  "public"."User"."name"
FROM "public"."User"
WHERE ("public"."User"."id") IN (
    SELECT "t0"."authorId"
    FROM "public"."Post" AS "t0"
    WHERE (
        "t0"."title"::text LIKE 'a'
        AND "t0"."content"::text LIKE 'b'
        AND "t0"."authorId" IS NOT NULL
      )
  )

More SQL improvement can be done by following these suggestions (Not implemented in this PR)

• Investigate Generated SQL for RelationFilters #1075
• Prisma generates unnecessary subquery, which is inefficient prisma#17879

[codspeed-hq]

CodSpeed Performance Report

Merging #3882 KhooHaoYit:main (7458eac) will not alter performances.

Summary

🔥 0 improvements
❌ 0 regressions
✅ 6 untouched benchmarks

🆕 0 new benchmarks
⁉️ 0 dropped benchmarks

[janpio]

(Rerunning hopefully flaky test)

[Weakky]

Hey @KhooHaoYit, thanks a lot for your second contribution! We'll make sure to review your PR soon 🙏

[Weakky]

Apologies for the late answer again @KhooHaoYit. As we're improving relational filters (mostly using joins for to-one relational filters), your PR will be a great additional improvement! Great job spotting it and thank you for the contribution 💯🙏.