tests: helpers: use less random data
We do not need really random data in the files we generate - this only slows
the tests down a lot. Let's fill the files with semi-random data instead - just
pick a random byte and fill large regions with this byt. This also make
compression work faster, and it is good enough for our purposes.
Besides, this allows to reproduce tests by setting a known seed, which does not
work with os.urandom.
Change-Id: I0459ee44d271b8e19581e84cc33b7167e4ada166
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@intel.com>