Reload manifest in RPC server when a SIGHUP signal is received

[drewbanin]

Describe the feature

When a sighup is received, the RPC server should reload the manifest in the parent process. New requests to the RPC server should use the fresh manifest when it is loaded.

Queries to the RPC server during a reload can either see the previous version of the manifest, or they can block until the manifest is loaded.

This is a sensible approach because the RPC servers uses subprocesses for each request and it's easier to pass state "down" to a subprocess than it is to pass it back "up" to the parent process. The alternative here would be to expose and endpoint like /reload, but I think a hup signal accomplishes the same thing while being simpler to implement.

[beckjake]

Fixed in #1699