context/cancel mechanism for RPC

In the work being done for CSI (in particular #6903) we've found that there are RPC calls that can block for very long periods of time, which in busy clusters with lots of volumes could lead to resource pressure on goroutines. It might be helpful to have a way to pass a cancellation context to RPC calls so they can be timed out.

We're not considering this blocking for CSI and it's probably not exclusive to CSI, so deserves a little more consideration than something we'd stick into the CSI Controller RPCs.

Review the RPC work described in #6903 and identify where long-running calls are defined and invoked; the issue names no files or tests. Compare the CSI-specific context with a broader RPC mechanism, and define done as an agreed approach for passing cancellation and timing out calls beyond the CSI Controller RPCs.

Written by the indexing model from the issue text.