Incorrect order of rpc invocations

When a paraview client invokes a bunch of RPCs in the thallium backend with arguments that differ greatly in bytesize, the server-side execution of those RPCs sometimes happens in a different order. This issue currently blocks !295 (merged). A unit-test that opens a group of data files tries to send a large state update, immediately followed by an information update. What really happens is that server sees the request to update information before the state update arrives. As the remote IOSSReader object is not yet created, the reader proxy fails to update information.

Are we missing crucial steps during thallium initialization? Should we use a custom scheduler? A minimal, reproducible example lives at https://github.com/jspanchu/thallium-rpc-ordering. The RPC pool, xstream and progress pool, xstream is setup exactly the way ParaView does it.