I did a quick test (with CUPS 1.5.4):

I made /usr/lib/cups/backend/myusb
to simulate such device discovery information:
-----------------------------------------------------------------------------
#! /bin/bash
# Output "device discovery" information on stdout:
if test "$#" = "0"
then echo 'direct usb "HP LaserJet 9999" "HP LaserJet 9999 needs HPLIP installed to work on USB"'
fi
exit 0
-----------------------------------------------------------------------------

Accordingly I get
-----------------------------------------------------------------------------
# lpinfo -l -v | grep -2 9999
Device: uri = usb
        class = direct
        info = HP LaserJet 9999 needs HPLIP installed to work on USB
        make-and-model = HP LaserJet 9999
        device-id = 
        location = 
-----------------------------------------------------------------------------

Now it depends on what "not break existing printer setup tools" 
should mean.

The above device discovery information is valid
according to "man 7 backend" and therefore it cannot
break existing printer setup tools that work in compliance
with CUPS.

But lpadmin fails (I use CUPS 1.5.4):
-----------------------------------------------------------------------------
# lpadmin -p testy -v usb -E
lpadmin: Bad device-uri "usb".

# echo $?
1
-----------------------------------------------------------------------------

I think it is perfectly right that lpadmin and all
printer setup tools that work in compliance with CUPS
fail to set up a print queue because such a print queue
cannot work and accordingly the user should get
a failure error message when he did not notice the
"HP LaserJet 9999 needs HPLIP installed to work on USB"
information or when a printer setup tool does not show
the backend's device-info output.

Currently the YaST2 printer setup tool does intentionally
not show device discovery information when no device-uri
but only the scheme was ouptut by a backend
(i.e. only "usb" instead of "usb:/...").

I described the reason in this comment in the
YaST2 printer setup tool sources:
-----------------------------------------------------------------------------
// do a special filtering for those connections which are supposed to work
// i.e. where the uri seems to be a complete DeviceURI e.g. "socket://192.168.1.2:9100"
// and not just an URI scheme like a plain "socket".
-----------------------------------------------------------------------------

The root cause is that lpadmin fails (at least up to CUPS 1.5.4)
when only the scheme but no real device-uri is provided:
-----------------------------------------------------------------------------
# lpinfo -v | grep -v :
network smb
network http
network ipp
network ipps
network lpd
direct usb
direct pipe
direct hal
network https
network beh
network socket
direct hp
direct hpfax

# for s in $( lpinfo -v | grep -v : | cut -d ' ' -f2 ) ; \
 do lpadmin -p testy -v $s -E ; echo $? ; done
lpadmin: Bad device-uri "smb".
1
lpadmin: Bad device-uri "http".
1
lpadmin: Bad device-uri "ipp".
1
lpadmin: Bad device-uri "ipps".
1
lpadmin: Bad device-uri "lpd".
1
lpadmin: Bad device-uri "usb".
1
lpadmin: Bad device-uri "hal".
1
lpadmin: Bad device-uri "pipe".
1
lpadmin: Bad device-uri "https".
1
lpadmin: Bad device-uri "beh".
1
lpadmin: Bad device-uri "socket".
1
lpadmin: Bad device-uri "hp".
1
lpadmin: Bad device-uri "hpfax".
1
-----------------------------------------------------------------------------

I do not want to show useless device discovery information
in the YaST2 printer setup tool
when queue setup cannot work with it.

I will re-think about it when special
device discovery information like
'direct usb "HP LaserJet 9999" "HP LaserJet 9999 needs HPLIP installed to work on USB"'
will be shown that is useful regardless
that queue setup cannot work with it.

It would be perfect if a printer setup tool could distinguish
such meaningful special device discovery information like
'direct usb "HP LaserJet 9999" "HP LaserJet 9999 needs HPLIP installed to work on USB"'
from arbitrary generic useless fallback device discovery information
as I currently get at least up to CUPS 1.5.4:
-----------------------------------------------------------------------------
# lpinfo -v | grep -v :
network smb
network http
network ipp
network ipps
network lpd
direct usb
direct pipe
direct hal
network https
network beh
network socket
direct hp
direct hpfax
-----------------------------------------------------------------------------

What could be used to distinguish meaningful device
discovery information from useless fallback device discovery
information is whether or not "device-make-and-model" and
"device-info" are non-empty strings (cf. "man 7 backend").