[atlas] Ability to traceroute to 240/4 - software vs. hardware probes

Hi!

I noticed that Phillip Homburg merged a RIPE Atlas client patch in
November of last year, based on our request, that removes the Atlas
probe's inability to test addresses in 0/8 and 240/4.

https://github.com/RIPE-NCC/ripe-atlas-probe-measurements/commit/59475f848b2e55fa14f6c43e62557b97ed005241

So, I just tried a couple of measurements to 240.1.2.3 to see how Atlas
would handle them:

https://atlas.ripe.net/measurements/48100582/
https://atlas.ripe.net/measurements/48100585/

(I don't expect these to succeed, as no part of 240/4 is publicly
announced in BGP; the interesting information to be garnered from this
isn't whether the probes succeed, but how they fail.)

For the first measurement, with ping, 6 hardware probes and 4 software
probes all gave the same result: "Unreachable" (which is what you would
get if you ran this on any ordinary Internet host with command-line
ping).

For the second measurement, with traceroute, 6 hardware probes all
actually attempted the traceroute (getting failures after some small
number of network hops), while 3 software probes all reported back
"!address not allowed".

What could account for the difference in how hardware and software
probes handle this task?  I would have thought that the change from
last November would make software probes willing to attempt these.