Adds a test case for when the following conditions cause a crash:
- a subsequent fragment is received before the first
- the reassembly buffer is currently filled up when another fragment of
a different datagram arrives and thus needs to be cached out to make
room for the new reassembly
While looking at tests/gnrc_ipv6_ext_frag again while writing
RIOT-OS/Release-Specs#137, I noticed that several of tests that I
definitely wrote myself from scretch are attributed wrong (and
sometimes even documented wrong). I guess this was caused by just
copy-pasting the files...
