Comment 1 for bug 1716706

We should be able to easily count build_requests in the API DB and include those in the count for instances, but the API also checks cores/ram when counting instances and we don't have that information in the build_request (well, we do, but it's in the serialized build_requests.instance field, which can't be queried via SQL).

And we can't use Placement to count VCPU/MEMORY_MB during the API check since the allocations aren't created in Placement yet (those get created when the scheduler picks a host).

So we might need to add vcpus/memory_mb fields on the build_requests table to mimic the instances table for counting those resources before the API casts to conductor.