[red-knot] mypy_primer: larger depot runner (#17547)
## Summary
A switch from 16 to 32 cores reduces the `mypy_primer` CI time from
3.5-4 min to 2.5-3 min. There's also a 64-core runner, but the 4 min ->
3 min change when doubling the cores once does suggest that it doesn't
parallelize *this* well.
