fix: client handling of batched queries when some fail, and some not

[miguelff]

Addresses prisma/prisma-engines#3458 (comment)

Depends on merging and updating engine hashes for prisma/prisma-engines#3458 as I need to write some unit tests to prevent regressions.

Before the release of native findXOrThrow, there were no instances of batching where commonly some promises failed while others succeeded. And the code handling that scenario was broken. This PR attempts to fix it.

[miguelff]

The regression affected binary and library engines. For data proxy, there's a different problem that manifests when executing the same test (The test exercises two different problems)

The problem in particular (after debugging the data proxy) is that the engine adapter is not able to discern errors coming from individual queries in a batch, and as a consequence, doesn't reject a promise that errored.

If we do this:

    const found = prisma.user.findUniqueOrThrow({ where: { id } })
    const notFound = prisma.user.findUniqueOrThrow({ where: { id: missing } })
    const result = await Promise.allSettled([found, notFound])

We get:

    [
      { status: 'fulfilled', value: { id: id } },
      { reason: new NotFoundError('No User found'), status: 'rejected' },
    ]

while with the data proxy, we get.

    [
      { status: 'fulfilled', value: { id: 'xqwhv5rgkt2' } },
      { status: 'fulfilled', value: undefined }
    ]

I leave to @aqrln the investigation, ticket creation, and (if doable directly) addressing of the problem described here. To reproduce it, it is enough to unskip the test provided in this patch.