QE: Optimize the compound id select

[pimeys]

In QE when using compound ids, the related fields select does something following:

WHERE ("id1" = 1 AND "id2" = 2) OR ("id1" = 3 AND "id2" = 4)

to search related records for given parent projections. This is very inefficient and will skyrocket the response times when having a larger set of projections. For now we've been optimizing the special case with just a single key with an IN statement that is orders of magnitude faster:

WHERE "id1" IN (1, 3)

We can do this with compound keys too, taking into account a slight difference how SQLite handles them, so for MySQL and PostgreSQL this will work:

WHERE ("id1", "id2") IN ((1, 2), (3, 4))

But for SQLite we need

WHERE ("id1", "id2") IN (VALUES (1, 2), (3, 4))

Implement optimizations for the compound ids and measure the effect with Chihiro.

[pimeys]

QE: #569
Quaint: prisma/quaint#92

[pimeys]

Done and merged.