Getting this wrong was where my previous series of commits went astray.

...to isolate them from the core logic. Yes, I got this wrong in the
reverted patch series.

In the past, we told reference clocks from real peers by giving the
reference clocks an address of the form 127.127.t.u, where t is the
type and u is the unit number.  In ntpd itself, the filtering that
used to be done based on this magic address prefix is now done
using the is_refclock_packet() test on incoming packets.  The
remaining instances of magic-address testing are in the
configuration-language interpreter only and are used to prevent
inapropriate configuration commands from being applied to
refclock entries.  They'll go away when the configuration syntax
is redesigned.

In theory, therefore, it would now be possible for ntpd to use a
server with an address in the 127.127.t.u range.  In practice this
is probably a bad idea as it would confuse ntpq, which keeps some
of these prefix checks in order to be able to recognize clock packets
by address only (that being all it has to work with).

De-confusing ntpq will require some modifications to mode 6
response formats so that the response to a peer query conveys
*explicitly* whether it's a refclock.  Even so, legacy ntpq
instances will still be confused.

...when reporting refids.  This affects the rv report.

...assumes the magic-refclock-address convention.

...and don't export numeric type any more, as we're trying to make that
strictly an internal convention.

This replaces the filtering on magic address that used to happen in ntpq.
It has the minor downside that a new ntpq displaying a peer spreadsheet
for an old ntpd will show refclocks by 127.127.t.u address.

This function is no longer needed, because shortname information for
clock drivers is now kept in the driver method table and exported
to ntpq via the displayname peer variable. This change eliminates
the last magic-address-mask test outside the configuration-file
interpreter.

Knowledge of the magic 127.127.t.u address convention for refclocks is
now confined to ntp_config.c and used only for interpreting
configuration syntax.

In theory we could now assign a refclock any unique IP address we
like; neither ntpd nor ntpq would care any more. In practice it's
probably best to continue assigning only 127.127.t.u addresses so
that (at least for a while) we refrain from confusing legacy ntpq
versions.

Together with the elimination of refnumtoa(), this sets the stage
for entirely removing the external visibility of numeric driver
types, and a better clock-configuration syntax.