I'm in-between on whether this is desired or not, but I suppose it is. It seems weird to not use the command context directly, but then the caller (like our tests) is responsible for canceling it, which is even weirder I suppose.

We're still inheriting it, so in that sense we're still using it. I think this change can make more sense if we consider other scenarios, like adding tracing to the commands. We'd want to ensure that the context we're passing along contains the tracing information which means we shouldn't be accessing the command context directly.