Joshua valgrind package should not run downgrade tests

[sfc-gh-anoyes]

Restarting tests are designed to see what happens if we start a cluster with the on-disk state from a different fdbserver version. These tests involve an "old binary" , which is not under test (and not valgrind-aware i.e. built with -DUSE_VALGRIND=ON), and a "new binary", which is under test. For upgrade tests, we run the binary under test under valgrind with state written by the old binary. This makes sense as reading the on-disk state could trigger a memory error.

For downgrade tests, we run the binary under test first, and then use the "old binary" only to detect issues with the on-disk state written by the binary under test. It's not necessary to run valgrind on the binary under test, since it's not doing anything that other workloads don't do, and it doesn't make sense to run valgrind on the "old binary" since it's not valgrind-aware and not under test.