Vectors (for pgvector) are not treated correctly

[MischaPanch]

Bug report

Describe the bug

When using a schema with vectors using the pgvector extension, the resulting SQL is syntactically incorrect.

To Reproduce

Setup a prisma schema like this

generator client {
  provider        = "prisma-client-js"
  previewFeatures = ["postgresqlExtensions"]
}

datasource db {
  provider   = "postgresql"
  url        = env("DATABASE_URL")
  directUrl  = env("DIRECT_URL")
  extensions = [pgvector(map: "vector", schema: "extensions")]
}

model MyEmbedding {
  id                 String                      @id @default(uuid()) @db.Uuid
  embedding          Unsupported("vector(8)")
  @@map["my_embeddings"]
}

Then creating entries with seed.myEmbedding will create invalid SQL. If an array [num1, num2, ...] is passed as value for embedding, the result is

INSERT INTO ... VALUES (<id>, num1, num2, ...)

The correct syntax is

INSERT INTO ... VALUES (<id>, ARRAY[num1, num2, ...]::vector)

I tried passing the string "ARRAY[num1, num2, ...]::vector" as value for embeddings. The sql then becomes

INSERT INTO ... VALUES (<id>, 'ARRAY[num1, num2, ...]::vector')

so it's almost correct, but not quite.

Expected behavior

The correct sql code is generated when vectors are passed, or some workaround for vectors is possible

[MischaPanch]

Maybe somewhat related to prisma/prisma#18442