async_rpc errors should contain targeting information at HostAndPort granularity

The async_rpc responses today contain the HostAndPort a command was run on only if it completed successfully. We also need to include this information in error cases - if we ever attempted to send a command to a HostAndPort, API users may want to send cleanup work to that particular node in the future, depending on the command (i.e. killCursors/_killOperations). 

Note that we should probably do some of this cleanup automatically in cases where we know it should be done - but this might require design of a more robust _killOperation operation. To re-host existing components, such as the ARS, that do their own cleanup, the API should expose this information for now.