Test more pack sizes in the AT and/or nightlies.

I tried to run `shoc_standalone` through valgrind, when building with pack size 8. It passed, but raised new valgrind errors (invalid read). Given that the test passed, I speculate it is a harmless error, but nevertheless, using a single pack size could hide bugs.

To be fair, we do test pack size 1 and 16, but they do not cover the whole span of scenarios. More precisely, we want to test the following;

a1) `npacks(nlev)==npacks(nlev+1)`
a2) `npacks(nlev)1

With pack sizes 16 (dbg and sp builds, in test-all-scream) and 1 (fpe build), and a test with 72 levs, we test a1+b2 and a2+b1, but the other two combinations are not tested. Developing Hommexx, we realized that there are _several_ bugs that can sneak in when we don't stress test the pb.

It might seem like there are 4 combinations of the above options, but it's clear that a1+b1 is not possible. But if we tested, say, pack size=8 with nlev=72, we'd also test a2+b2.

Long story short, when out pack size>1, for a given test we are only testing either a1 or a2, but that might hide some bugs. We need to test both.