AFAICT, this has been mildly broken forever.

If you look at store_register_using_P and fetch_register_using_p,
they both call remote_send, followed by packet_ok, and then send
error if PACKET_ERROR.

Unfortunately the PACKET_ERROR path is dead code, because remote_send
already bails with a much more general (ie. less useful) error msg
if we get back an Exx reply.

Most of us will have never seen this happen, because how can a
remote target fail to write a register?

But it can happen when the target is a replay recording as with
Virtutech and VMware targets.

This change is to give a more helpful error message, eg.
'Could not write register "eip"', as opposed to
'Remote failure reply: E00'.