Per the SDAM rules in the error handling section, a non-timeout
error encountered during operation execution MUST transition the
server to an `Unknown` state and kick off the SDAM flow.

NODE-2435

If there was enough data buffered to read multiple messages, the
MessageStream would currently callback to the underlaying Writable
callback multiple times. Also fixed here is that the callback would
not be called in the event that there wasn't enough data to process
a single message

NODE-2437

This work acknowledges an `exhaustAllowed` option to enable exhaust
behavior on OP_MSG messags. It also tracks synthetic `getMore` ids
and uses the same callback provided for the orignal command for all
returned batches.

NODE-2438

This refactoring improves readability and reduces coupling between
the `Server` and `Topology` types. Previously errors experienced
at the `Server` level would emit an `error` event, which needed to
be caught by the `Topology`. Since nothing about error handling
required access to the `Topology` type, all of that behavior was
moved back into the `Server` type.

NODE-2449