Read: Making zeroing of buffer configurable

It is only relevant for sparse files but the current implementation
has a too high overhead on a single node (~10%). We need to find
a better solution at a later point.