Investigate whether RPCs can use manual object quarantine directory

In order to speed up access checks, we can reuse information from quarantine directories to avoid scaling with repo size and instead scale with change size. While this works perfectly for pushes, it doesn't in the context of RPCs: we don't create object quarantine directories there, and as such we cannot use this important optimization.

We should check how involved it is to start using object quarantine directories in our RPCs. This will buy us a bunch of benefits:

• concurrent RPCs don't influence each other because they don't see newly created objects of others
• we can easily discard objects in case the RPC fails
• perfomance in access checks is going to be a lot better