Fix 500 error when we get an error handling a PDU

[richvdh]

FederationServer doesn't have a send_failure (and nor does its subclass,
ReplicationLayer), so this was failing.

I'm not really sure what the idea behind send_failure is, given (a) we don't do
anything at the other end with it except log it, and (b) we also send back the
failure via the transaction response. I suspect there's a whole lot of dead
code around it, but for now I'm just removing the broken bit.

[erikjohnston]

Since we're already sending the error back in the response, its probably fine.

The failure mechanism is/was meant to be a way to asynchronously tell the sending server they're a crank, e.g. if we stuck incoming federation events into a queue rather than doing the full auth checks up front.

[richvdh]

e.g. if we stuck incoming federation events into a queue rather than doing the full auth checks up front.

Oh, like we do during room joins (IIRC). I suppose we should wire that up sometime.